Electrical Vestibular Stimulation forParkinson’s Disease TreatmentbySoojin LeeB.Sc., Korea University, 2008M.Sc., The University of British Columbia, 2010A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Biomedical Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)July 2019c© Soojin Lee 2019The following individuals certify that they have read, and recommend tothe Faculty of Graduate and Postdoctoral Studies for acceptance, the thesisentitled:Electrical Vestibular Stimulation for Parkinson’s Disease Treat-mentsubmitted by Soojin Lee in partial fulfillment of the requirements for thedegree of Doctor of Philosophy in Biomedical Engineering.Examining Committee:Martin J. McKeown, Faculty of Medicine (Neurology)SupervisorZ. Jane Wang, Electrical and Computer EngineeringCo-supervisorSilke Cresswell, Faculty of Medicine (Neurology)Supervisory Committee MemberPurang Abolmaesumi, Electrical and Computer EngineeringUniversity ExaminerTodd Woodward, Faculty of Medicine (Psychiatry)University ExaminerHong Bo, Biomedical Engineering, Tsinghua UniveristyExternal ExaminerAdditional Supervisory Committee Members:Cyril Leung, Electrical and Computer EngineeringSupervisory Committee MemberiiAbstractParkinsons disease (PD) is a progressive movement disorder characterizedby degeneration of dopaminergic neurons and abnormal brain oscillations.While invasive deep brain stimulation can improve some motor deficits bydisrupting pathological brain oscillations, achieving comparable results withnon-invasive brain stimulation (NIBS) remains elusive. Previous studieshave suggested that electrical vestibular stimulation (EVS) may amelioratesome motor symptoms in PD. However, the investigated effects are limited toa few domains, only a handful of stimulation waveforms have been explored,and neuroimaging studies capable of probing the mechanisms are greatlylacking. The overarching objective of this thesis is to utilize biomedicalengineering approaches to fully explore the EVS technique as a potentialtherapeutic intervention for PD. This involves development of new stimuli,development of new artifact rejection methods, and thorough investigationsof brain and behavioural responses, as outlined below.To achieve the objective, noisy EVS is firstly revisited and tested withPD and healthy subjects to investigate effects on visuomotor tracking be-haviours. Next, novel EVS stimuli are developed using multisine signals indistinct frequency bands and tested in the experiment where the stimuli areapplied to PD and healthy subjects during rest and task conditions whileEEG are being recorded. This simultaneous EVS-EEG study aims to pro-vide insights into modulatory effects of EVS on brain oscillations and motorbehaviours altered in PD and whether the effects are a function of differentstimulation types. One critical challenge involved with EVS-EEG studiesis that EEG recordings are severely corrupted by the stimulation artifacts.To resolve this, a quadrature regression and subsequent independent vec-tor analysis method is developed and its superior denoising performance toconventional methods is demonstrated. Finally, underlying mechanisms ofEVS effects in PD are investigated in a resting-state functional MRI study.The results from this thesis suggest that sub-threshold EVS in PD in-duces widespread motor changes and brain activities that are stimulus-dependent, suggesting subject-specific stimuli may ultimately be desirableto achieve a clinically meaningful effect.iiiLay SummaryPeople with Parkinsons disease (PD) experience debilitating motor symp-toms, which are associated with abnormal brain activities. Current treat-ments include medication and invasive surgical implantation of electrodesinto a deep region of the brain to deliver electrical impulses. There is keen in-terest in finding ways to non-invasively stimulate the brain safely to treatedPD. Electrical vestibular stimulation (EVS) is a non-invasive brain stimu-lation technique that delivers weak electrical currents to the balance organlocated in the inner ear, and also induces changes in brain activities. Thepurpose of this dissertation is to understand how it might go about relievingmotor symptoms in PD and provide a deeper understanding of how EVSworks. In addition, to further our understanding, this thesis demonstratesthat customizing the stimulation to the individual may be necessary.ivPrefaceThis dissertation is primarily based on five journal publications (three areunder review), one conference paper and nine conference presentations, re-sulting from a collaboration between multiple researchers. These publi-cations have been modified to make the dissertation coherent. The au-thor was responsible for design of experiment and data collection. Partici-pants recruitment and scheduling was assisted by Ms. Christina Jones andMs. Tammy Kang. The author was also responsible for the data analyses,evaluation of the methods and the production of the manuscripts. All co-authors have contributed to the editing of the manuscripts and providingfeedback and comments. The dissertation work was conducted in UBCsPacific Parkinson Research Centre. Approval of this study was obtained bythe UBCs Clinical Ethics Board (Certificate number: H09-02016).The study from Chapter 2 is based on:• Soojin Lee, Diana J. Kim, Daniel Svenkeson, Gabriel Parras, MeekoMitsuko K. Oishi and Martin J. McKeown. Multifaceted effects ofnoisy galvanic vestibular stimulation on manual tracking behavior inParkinsons disease. Frontiers in Systems Neuroscience, 9(5), 1-9, 2015.and was presented at:• Soojin Lee, Daniel Svenkeson, Meeko Oishi and Martin J. McKeown.Multifaceted effects of noisy galvanic vestibular stimulation on manualtracking behavior in Parkinsons disease. IEEE EMBS BRAIN GrandChallenge, November 2014, Washington, USA.• Soojin Lee, Diana Kim, Daniel Svenkeson, Meeko M.K. Oishi and Mar-tin J. McKeown. Synergistic effects of noisy galvanic vestibular stim-ulation and oral L-dopa in improving manual tracking performance inParkinsons disease. 1st International Brain Stimulation Conference,March 2015, Singapore.vPrefaceThe contribution of the author was the data analyses and writing themanuscript. Diana Kim contributed to collecting the data and editing themanuscripts. Dr. Meeko Oishi, Daniel Svenkeson and Gabriel Parras pro-vided valuable feedback on the data analyses.The study from Chapter 3 is based on:• Soojin Lee, Z. Jane Wang, Martin J. McKeown, and Xun Chen, Re-moval of High-Voltage Brain Stimulation Artifacts from Simultane-ous EEG Recordings. IEEE Transactions on Biomedical Engineering,66(1), 50-60, 2019.and was presented at:• Soojin Lee, Z. Jane Wang and Martin J. McKeown, Removal of high-voltage brain stimulation artifacts from simultaneous EEG record-ings. 2nd International Brain Stimulation Conference, March 2017,Barcelona, Spain.The contribution of the author was the data collection, the analysesof the data and writing the manuscript. Dr. Jane Wang contributed toprovided valuable scientific inputs to improve the proposed method andfeedback on the data analyses.The study from Chapter 4 is based on:• Soojin Lee, Aiping Liu, Z. Jane Wang, and Martin J. McKeown.,Abnormalphase coupling in Parkinsons disease and normalization effects of sub-threshold vestibular stimulation. Frontiers in Human Neuroscience,13(118), 1-15, 2019.The contribution of the author was the development of the EVS stimuli,design of the experiment, the data collection and the analyses of the data,and writing the manuscript. Dr. Aiping Liu and Dr. Jane Wang providedvaluable feedback on the data analyses.The study from Chapter 5 is based on:• Soojin Lee, Z. Jane Wang, and Martin J. McKeown, High-frequencyvestibular simulation jointly affects motor performance and beta os-cillations in Parkinson’s disease. (under review).viPrefaceand was presented at:• Soojin Lee, Z. Jane Wang and Martin J. McKeown, Engineering Ap-proaches to Non-Invasive Electrical Stimulation of the Brain: Appli-cation to Parkinsons Disease. Emerging Technologies 2018, May 2018,Whistler, Canada.• Soojin Lee, Z. Jane Wang and Martin J. McKeown, Non-Invasive Gal-vanic Vestibular Stimulation Augments Event-Related Desynchroniza-tion and Improves Motor Performance in Parkinsons Disease. Cana-dian Student Health Research Forum, June 2018, Winnipeg, Canada• Soojin Lee, Z. Jane Wang and Martin J. McKeown, Non-Invasive gal-vanic vestibular stimulation augments beta desynchronization and im-proves motor performance in Parkinsons Disease. 3nd InternationalBrain Stimulation Conference, February 2019, Vancouver, Canada.The contribution of the author was the development of the EVS stimuli,design of the experiment, the data collection, the analyses of the data, andwriting the manuscript. Dr. Jane Wang provided valuable scientific inputfor the data analyses.The study from Chapter 6 is based on:• Soojin Lee, Aiping Liu, Jowon L. Kim, Saurabh Garg, Z. Jane Wang,and Martin J. McKeown. Effects of electrical vestibular stimulationon thalamic activity and disrupted basal ganglia-thalamic connectivityin Parkinsons disease. Parkinsonism and Related Disorders (underreview).The contribution of the author was the data collection, the analyses ofthe data, and writing the manuscript. Jowon Kim contributed to the datacollection, and Saurabh Garg contributed to preprocessing of the fMRI dataand interpretation of the data. Dr. Aiping Liu contributed to preprocessingof the fMRI data, performance of normalized cut spectral clustering analysis,and interpretation of the results. Dr. Jane Wang provided feedback andtechnical input to the data analyses.Finally, my supervisor, Dr. Martin McKeown, helped with the devel-opment, implementation and evaluation of all of the methods in the aboveviiPrefacepublications and this dissertation. He also significantly contributed to bi-ological interpretation of the results and improvement of the manuscriptsthrough valuable comments and feedback.viiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xivList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xviiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Parkinson’s Disease (PD) . . . . . . . . . . . . . . . . . . . . 11.1.1 Brief History, Epidemiology and Clinical Features . . 11.1.2 Neuropathology . . . . . . . . . . . . . . . . . . . . . 21.1.3 Aberrant Neural Oscillations in PD . . . . . . . . . . 31.1.4 Current Treatment Options for PD . . . . . . . . . . 51.2 Non-invasive Brain Stimulation (NIBS) . . . . . . . . . . . . 61.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . 61.2.2 Stimulation Parameters and Protocols . . . . . . . . . 71.2.3 Proposed mechanisms of tES . . . . . . . . . . . . . . 81.2.4 Clinical Research Findings on PD using tES . . . . . 101.2.5 Challenges and Open Questions . . . . . . . . . . . . 101.3 Electrical Vestibular Stimulation (EVS) . . . . . . . . . . . . 131.3.1 Background . . . . . . . . . . . . . . . . . . . . . . . 131.3.2 The Vestibular System . . . . . . . . . . . . . . . . . 14ixTable of Contents1.3.3 EVS Effects on PD . . . . . . . . . . . . . . . . . . . 151.3.4 Stimulation Parameters . . . . . . . . . . . . . . . . . 151.3.5 Multisine Signal . . . . . . . . . . . . . . . . . . . . . 161.4 Research Objectives and Thesis Outline . . . . . . . . . . . . 201.4.1 Research objectives . . . . . . . . . . . . . . . . . . . 201.4.2 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . 222 Discriminant Feature Detection in Manual Tracking Behavioursin PD and Effects of Noisy EVS . . . . . . . . . . . . . . . . 262.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 282.2.1 Subjects . . . . . . . . . . . . . . . . . . . . . . . . . 282.2.2 Ethics Statement . . . . . . . . . . . . . . . . . . . . 292.2.3 Visuomotor Tracking Task . . . . . . . . . . . . . . . 292.2.4 Stimulus . . . . . . . . . . . . . . . . . . . . . . . . . 302.2.5 Behavioural Data Analysis . . . . . . . . . . . . . . . 312.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.3.1 Results of LDA in Worse Condition . . . . . . . . . . 342.3.2 Results of LDA in Better Condition . . . . . . . . . . 342.3.3 Results of Robust Regression Model . . . . . . . . . . 352.3.4 Effect of EVS on Cursor Overshooting . . . . . . . . 352.3.5 Effect of EVS on SNR of Cursor Trajectory . . . . . 372.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383 Quadrature regression and IVA approach to removal of high-voltage EVS artifacts from simultaneous EEG recordings 413.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 413.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 443.2.1 Subjects . . . . . . . . . . . . . . . . . . . . . . . . . 443.2.2 EVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.2.3 Study Protocol . . . . . . . . . . . . . . . . . . . . . . 453.2.4 EEG Data Acquisition and Preprocessing . . . . . . . 473.2.5 Simulation Data . . . . . . . . . . . . . . . . . . . . . 473.2.6 Artifact Rejection Methods . . . . . . . . . . . . . . . 503.2.7 Performance Evaluation . . . . . . . . . . . . . . . . . 523.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.3.1 Simulation Results . . . . . . . . . . . . . . . . . . . . 553.3.2 Real Data Results . . . . . . . . . . . . . . . . . . . . 583.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613.4.1 Advantages of JBSS Approaches . . . . . . . . . . . . 61xTable of Contents3.4.2 Recommendations and Limitations . . . . . . . . . . 634 Sparse Discriminant Analysis for Detection of PathologicalDynamic Features of Cortical Phase Synchronizations in PD 654.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 684.2.1 Participants . . . . . . . . . . . . . . . . . . . . . . . 684.2.2 EVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 694.2.3 EEG recording . . . . . . . . . . . . . . . . . . . . . . 714.2.4 EEG preprocessing . . . . . . . . . . . . . . . . . . . 714.2.5 Phase Locking Value (PLV) . . . . . . . . . . . . . . 714.2.6 Sparce Discriminant Analysis . . . . . . . . . . . . . . 724.2.7 Statistical Analysis . . . . . . . . . . . . . . . . . . . 734.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.3.1 SDA Classification Results and Selected Features . . 734.3.2 Group Comparison of Baseline PLV Features . . . . . 744.3.3 Online- and after-effects of EVS . . . . . . . . . . . . 764.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784.4.1 Disrupted Cortical Coupling Strength in the MotorRegions . . . . . . . . . . . . . . . . . . . . . . . . . . 794.4.2 Variablity and Entropy of PLV in the Theta Band . . 814.4.3 Variability and Entropy of PLV in the Alpha Band . 814.4.4 PLV sample entropy is higher in the long-range gammaactivity in PD . . . . . . . . . . . . . . . . . . . . . . 824.4.5 Normalizing Effects of EVS and Potential Mechanisms 834.4.6 Limitations . . . . . . . . . . . . . . . . . . . . . . . . 845 Discriminant Correlation Approach to Joint Estimation ofMaximal EVS Effects on Motor Behaviour and Cortical BetaOscillations in PD . . . . . . . . . . . . . . . . . . . . . . . . . 865.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 865.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 885.2.1 Subjects . . . . . . . . . . . . . . . . . . . . . . . . . 885.2.2 EVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 885.2.3 Study protocol . . . . . . . . . . . . . . . . . . . . . . 895.2.4 EEG Recordings and Preprocessing . . . . . . . . . . 905.2.5 Data analysis . . . . . . . . . . . . . . . . . . . . . . 905.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 935.3.1 Task Performance in the Sham Condition . . . . . . . 935.3.2 EVS Effects on the Task Performance . . . . . . . . . 93xiTable of Contents5.3.3 EVS Effects during Motor Preparation . . . . . . . . 955.3.4 EVS Effects during Motor Execution . . . . . . . . . 975.3.5 EVS Effects on Temporal Patterns of the Beta ERD . 1005.3.6 EVS Effects on the Beta Oscillations in the RestingCondition . . . . . . . . . . . . . . . . . . . . . . . . 1005.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1005.4.1 Abnormalities in Motor Control in PD . . . . . . . . 1015.4.2 EVS effects on the behavioural indices . . . . . . . . 1035.4.3 Functional significance of beta ERD in voluntary move-ment . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035.4.4 Modulation of beta ERD via NIBS . . . . . . . . . . 1055.4.5 Beyond the modulation of cortical oscillations —po-tential mechanisms of EVS . . . . . . . . . . . . . . . 1056 Spectral Clustering and Discriminant Correlation Approachto Estimation of EVS Effects on Functional Thalamic Subre-gions and BG–thalamic Connectivity in PD—fMRI Study 1086.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 1086.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 1106.2.1 Subjects . . . . . . . . . . . . . . . . . . . . . . . . . 1106.2.2 EVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1116.2.3 MRI acquisition and preprocessing . . . . . . . . . . . 1126.2.4 Data preprocessing . . . . . . . . . . . . . . . . . . . 1126.2.5 Connectivity based parcellation of thalamus . . . . . 1136.2.6 Subregion size analysis . . . . . . . . . . . . . . . . . 1146.2.7 Thalamus-BG connectivity analysis . . . . . . . . . . 1146.2.8 Statistical analysis . . . . . . . . . . . . . . . . . . . . 1156.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1166.3.1 Thalamus Parcellation . . . . . . . . . . . . . . . . . 1166.3.2 Thalamic subregion sizes . . . . . . . . . . . . . . . . 1166.3.3 Connectivity between the left BG and thalamic sub-regions . . . . . . . . . . . . . . . . . . . . . . . . . . 1176.3.4 Connectivity between the right BG and thalamus sub-regions . . . . . . . . . . . . . . . . . . . . . . . . . . 1206.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1206.4.1 Functional thalamic subregion sizes altered in PD . . 1206.4.2 Asymmetric connectivity of left BG and thalamus . . 1216.4.3 Potential mechanism of EVS . . . . . . . . . . . . . . 122xiiTable of Contents7 Conclusion and Future Work . . . . . . . . . . . . . . . . . . 1237.1 Conclusion and Summary . . . . . . . . . . . . . . . . . . . . 1237.1.1 Conclusion and Summary . . . . . . . . . . . . . . . . 1237.1.2 Limitations and suggestions for future work . . . . . 126Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129xiiiList of Tables1.1 A summary of tES research on PD . . . . . . . . . . . . . . . 111.2 Effects of varying EVS parameters on physiological responses 171.3 Seven multisine stimuli and corresponding frequency band-widths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.1 Demographical data of the PD subjects . . . . . . . . . . . . 282.2 Variables in LDA model . . . . . . . . . . . . . . . . . . . . . 322.3 Estimated coefficients in the robust regression model (Eq.2.2) and the P value . . . . . . . . . . . . . . . . . . . . . . . 332.4 Means of cursor overshooting on sinusoidal peaks and ANOVAresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373.1 Artifact Rejection methods used in EEG-NIBS Studies . . . . 433.2 Subjects information . . . . . . . . . . . . . . . . . . . . . . . 454.1 Demographic and clinical characteristics of the patients withParkinson’s disease (PD) and healhty controls (HC) . . . . . 685.1 Demographic and clinical characteristics of the patients withParkinsons disease (PD) and healthy controls (HC) . . . . . . 895.2 ANOVA results on the effects of stimulation condition on thebehaviour indices . . . . . . . . . . . . . . . . . . . . . . . . . 955.3 Linear regression analysis to demonstrate a relationship be-tween fast responses and vigorous movements. . . . . . . . . . 1025.4 Results of the repeated measures ANOVA to investigate ac-cumulated and/or learning effect on the behaviour measures. 1046.1 Study cohort demographics . . . . . . . . . . . . . . . . . . . 111xivList of Figures1.1 Simplified illustration of the main connections of the BG . . . 31.2 The spatial and temporal resolution at which different braininterventions work . . . . . . . . . . . . . . . . . . . . . . . . 71.3 Clipping algorithm for minimizing crest factor of multisinesignals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.1 Behaviour Task . . . . . . . . . . . . . . . . . . . . . . . . . . 292.2 Characteristics of the stimulus . . . . . . . . . . . . . . . . . 312.3 Coefficients of the variables of the linear discriminant functionin the Worse condition . . . . . . . . . . . . . . . . . . . . . . 342.4 Coefficients of the variables of the linear discriminant functionin the Better condition . . . . . . . . . . . . . . . . . . . . . . 352.5 Trajectories of target (blue) and cursor (EVSon: red, EVSoff:black) and ∆g (black bar in the bottom) . . . . . . . . . . . . 362.6 Representative example of cursor overshooting on upper andlower peaks from Subject 1 . . . . . . . . . . . . . . . . . . . 372.7 Comparison of SNR of cursor trajectories between EVSon andEVSoff conditions . . . . . . . . . . . . . . . . . . . . . . . . . 383.1 Experimental setup and an overall flow diagram for the study 463.2 The electrical circuit model for the physical electrode-skininterface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483.3 Example of generating simulated EEG data, Ys . . . . . . . . 493.4 The principal components (PC) from PCA, underlying sourcecomponents (IC) from SOBI, IVA and q-IVA, and canonicalvariates (CV) from MCCA . . . . . . . . . . . . . . . . . . . 533.5 Sample traces (8 channels) of the cleaned EEG data afterusing different artifact rejection methods . . . . . . . . . . . . 563.6 Comparison of the performance of different artifact rejectionmethods in the simulation study . . . . . . . . . . . . . . . . 573.7 Power spectrum of the channel O1 averaged over 10 subjects 59xvList of Figures3.8 The alpha activity at the channel O1 in the eyes-open (EO)and eyes-closed (EC) conditions . . . . . . . . . . . . . . . . . 604.1 The multisine stimuli and the phase locking value (PLV) cal-culation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704.2 Nonzero features selected by sparse discriminant analysis (SDA) 754.3 Group comparison of the discriminant component obtainedfrom the SDA and Pearson correlations with clinical scores . 764.4 Effects of EVS on the PLV mean . . . . . . . . . . . . . . . . 774.5 Effects of EVS on the PLV variability . . . . . . . . . . . . . 794.6 Effects of EVS on the PLV entropy . . . . . . . . . . . . . . . 805.1 Schematic of study protocol and motor task performance . . 915.2 Motor task performance in sham condition . . . . . . . . . . 945.3 EVS Effects on the motor task performance . . . . . . . . . . 965.4 DCA results demonstrating EVS1 effects on the beta ERDand task performance during motor preparation . . . . . . . . 975.5 DCA results demonstrating EVS1 effects on the beta ERDand task performance during motor execution . . . . . . . . . 985.6 DCA results demonstrating EVS2 effects on the beta ERDand task performance during motor execution . . . . . . . . . 995.7 Effects of EVS1 on the beta power during the 60-s restingcondition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1015.8 Effects of EVS2 on the beta power during the 60-s restingcondition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.9 Schematic representation of the major projections involvedwith motor functions . . . . . . . . . . . . . . . . . . . . . . . 1066.1 An example of thalamus parcellation results from a subject isdisplayed on the horizontal slices arranged from superior toinferior (from left to right) . . . . . . . . . . . . . . . . . . . . 1176.2 Comparison of the transformed subregional size obtained fromthe logistic regression and EVS effects . . . . . . . . . . . . . 1186.3 DCA results for the connectivity between left BG and bilat-eral thalami and EVS effects . . . . . . . . . . . . . . . . . . 119xviGlossaryBG Basal GangliaCCA Canonical Correlation AnalysisCF Crest FactorDBS Deep Brain StimulationDCA Discriminant Correlation AnalysisEEG ElectroencephalographyERD Event-Related DesynchronizationEVS Electrical Vestibular StimulationfMRI Functional Magnetic Resonance ImagingICA Independent Component AnalysisIVA Independent Vector AnalysisJBSS Joint Blind Source SeparationL-dopa LevodopaLDA Linear Discriminant AnalysisLFP Local Field PotentialMCCA Multiset Canonical Correlation AnalysisMEG MagnetoencephalographyNIBS Non-Invasive Brain StimulationPCA Principal Component AnalysisPD Parkinson’s DiseasePLV Phase Locking ValueSDA Sparse Discriminant AnalysisSNR Signal-to-Noise RatiotACS Transcranial Alternating Current StimulationtDCS Transcranial Direct Current StimulationtES Transcranial Electrical StimulationTMS Transcranial Magnetic StimulationtRNS Transcranial Random Noise StimulationUPDRS Unified Parkinson’s Disease Rating ScalexviiAcknowledgementsFirst and foremost, I would like to express my sincerest gratitude to Dr. Mar-tin McKeown, my supervisor, for the opportunity he gave me that openeda new chapter of my life. I am especially grateful for his great mentorship,invaluable support and trust that led me to be able to continuously growand become a better person who, like him, devotes talent and knowledge tocontributing to society. Thanks for the advice and generosity you offered meboth academically and personally throughout my experience at UBC. Thisthesis and my career about to start would never be possible without yoursupport.I am very grateful for my co-supervisor, Dr. Jane Wang, whose insightfuladvice helped me throughout the course of my PhD journey. Her genuinesupport and dedication were well beyond the role of co-supervisor, which Inot only appreciate but also admire.I would like to thank thesis examination committee members and chair,Drs. Abolmaesumi, Cresswell, Leung, Woodward and Shaw for their valu-able time, constructive comments and encouragement.Throughout the span of five years, I have been fortunate to work with andgetting to know many great individuals. I am thankful to all my lab mates,Saurabh Garg, Christy Jones, Stephanie Tran, Marcus Cheung, Maria Zhu,Robert Baumeister, Sue-Jin Lin, Jowon Laura Kim, Emma Kiss and SunNee Tan for creating supportive and fun environment. I also would like tothank Aiping Liu, Jiayue Cai, Azadeh Hosseini, and Xun Chen for theirinsightful advice and supports, Sangwook Han, Esther Song, Agnes Kwok,Inae Lee, and my office mates, Drs. Pratibha Surathi, Michele Matarazzo,Devavrat nene and Jose Wijnands for bringing me laughter and comfortthroughout this journey.To the research participants, thank you all for your enthusiasm andinterests in participating in my research.I am deeply grateful for my parents’ endless love, encouragement andsacrifice from when I was born till I grew up. I am happy you are my momand dad and you will be always in my heart throughout my life. My sisterand brother, Sooyoon and Yongmoon, thank you for always being supportivexviiiAcknowledgementsand unconditionally helping me out. A special thank you goes to Matthewfor his supports, love and humours that have strengthen me in many ways.Finally, this work was supported by PPRI/UBC chair in Parkinson’sDisease awarded to my supervisor and a generous gift from the MottersheadFoundation and PPRI.xixDedicationTo people with Parkinson’s disease— for generous support and great inspiration for this researchxxChapter 1Introduction1.1 Parkinson’s Disease (PD)1.1.1 Brief History, Epidemiology and Clinical FeaturesParkinsons disease (PD) is a progressive neurodegenerative disorder charac-terized by a large number of motor and non-motor features. In his landmarkpubliation in 1817 of “An essay on the shaking palsy”, James Parkinson firstdescribed the clinical syndrome that was later to bear his name [279]. About100 years passed (1919) after the first description of PD before it was rec-ognized that patients with PD prematurely lose cells in the substantia nigrapars compacta, and after 140 years had passed (1957) dopamine was dis-covered as a putative neurotransmitter [36, 160]. Later in 1961, the firsttrials of levodopa injection to improve akinesia in patients with PD wereconducted followed by the development of oral levodopa later in the decade[34, 73], which has remained the gold standard of treatment to date.PD is the second most common neurodegenerative disorder affecting 1–2% of people over age 65 years [335], with its prevalence escalating to ashigh as 4% with increasing age [396]. Worldwide incidence estimates of PDrange from 5 to > 35 new cases per 100,000 individuals yearly, dependingthe demographics of the populations studied [297]. The mean age of onsetis around 55 years old and the incidence increases 5–10-fold from the sixthto the ninth decade of life [334, 387, 396]. The number of people with PD isexpected to double between 2005 and 2030 according to recent meta analy-ses [152, 305], which is presumably due to growing elderly populations. Theprevalence of PD varies according to sex, race, ethnicity and environment.The incidence is greater in men in most populations, and African Ameri-cans and Asians may be less likely to be diagnosed with PD [78, 396], butit is difficult to determine the relative contribution of each of the factors.Mortality in PD increases to double compared to non-PD population afterthe first decade of disease onset [295], and mean PD duration until deathranges from 6.9 to 14.3 years [226].PD is most recognized for its cardinal motor symptoms including bradyki-11.1. Parkinson’s Disease (PD)nesia (slowness of movement), tremor, rigidity and postural instability andclinical diagnosis is defined by the presence of bradykinesia and rigidityand/or rest tremor. In addition to the motor features, a multitude of non-motor symptoms such as cognitive impairment (including executive dys-function, dementia, memory retrieval deficits and hallucination), autonomicdysfunction, disorders of sleep and depression are part of the disease [297].Progressive disease ultimately results in treatment-resistant motor symp-toms such as freezing of gait, falling and dyskinesia. PD evolves with differ-ent clinical courses and prognoses in individuals and thus it is increasinglyrecognized that PD is not a single entity but a heterogeneous disorder witha broad spectrum of motor and non-motor features [378].1.1.2 NeuropathologyPathologically, the disease is defined by the degeneration of dopaminergicneurons in the substantia nigra pars compacta (SNc) that project to thebasal ganglia (BG), a group of subcortical nuclei located deep within thebrain. The BG include the caudate, putamen, globus pallidus, the sub-stantia nigra and the subthalamic nucleus (STN) and are associated with avariety of functions, including control of voluntary motor movements. Thedegeneration of dopaminergic SNc neurons and their projections to the stria-tum may take decades to develop and recognizable motor or non-motor fea-tures appear only after substantial degeneration (∼60%) of the nigrostriatalneurons [102]. Earlier degeneration of SNc projections to the putamen thanthose to associative or limbic areas of the striatum may result in earlierdevelopment of the motor symptoms than the non-motor symptoms in PD[121]. The motor and non-motor symptoms in PD are multifactorial andalso linked to damage of specific brainstem nuclei [130]. The brainstem isdivided into mesencephalon, metenchephalon (pons), and medulla oblon-gata and includes the sensory and motor nuclei of 10 cranial nerves [170].The dorsal motor vagal nucleus, intermediate reticular zone, pedunculopo-nine nucleus are known to be particularly affected by PD and associatedwith gastrointestinal system dysfunction, pain, sleep disturbances, and gait[130].For decades, a functional and anatomical model of the BG circuitry hasbeen proposed to explain the clinical symptoms of PD (Fig. 1.1). Accord-ing to the model, the internal segment of the globus pallidus (GPi) receivessignals from the putamen through “direct” and “indirect” pathways. Asdopamine produced from the SNc modulates antagonistic functions in thedirect and indirect pathways, imbalanced activity between these two path-21.1. Parkinson’s Disease (PD)ways has been proposed to underlie the motor symptoms observed in PD[190]. However, recent data from different experimental approaches indi-cate that this model alone cannot explain many key features of the disease[37, 212]. For instance, it does not account for tremor and rigidity commonlyobserved in PD, and fails to explain why lesion treatments such as GPi pal-lidotomy paradoxically improve dyskinesias without any clear deleteriouseffects on motor function [47].Figure 1.1: Simplified illustration of the main connections of the BG. The direct andindirect pathways from the putamen have net effects of disinhibition and inhibition on thecortex, respectively. Reduced dopaminergic stimulation from SNc to the putamen in PDis marked with a black cross. Dopamine deficit leads to increased activity in the indirectpathway, in which STN hyperactivity is a key characteristic, and hypoactivity in thedirect circuit. Together, these alterations result in increased GPi/SNr output inhibitionof the thalamus and reduced activation of cortical and brainstem motor regions. Greenand red arrows denote excitatory and inhibitory activity, respectively. Figure modifiedfrom [317] and [435] (GPe: external globus pallidus; GPi: internal globus pallidus; SNc:substantia nigra; SNr: substantia nigra pars reticulta; STN: subthalamic nucleus; VTA:ventral tegmental area)Another characteristic feature of PD is abnormal accumulation of in-tracellular protein (α-synuclein) in widespread brain regions. Lewy bodies,fibrillary aggregates largely made up of α-synuclein, initially can be seenin neurons in the brainstem and olfactory system and are found in limbicand neocortical brain regions as the disease progresses [297]. The abnormalaggregation of α-synuclein are found in 10% of pigmented neurons in thesubstantia nigra and >50% in the locus ceruleus in PD [411].1.1.3 Aberrant Neural Oscillations in PDResearch has suggested that, in addition to the dopaminergic biochemicalchanges, aberrant neural synchrony is closely associated with manifestationof motor symptoms in PD. The functional role of the aberrant patterns31.1. Parkinson’s Disease (PD)of neural oscillatory activities in PD has been well investigated in stud-ies where local field potentials (LFPs) were recorded from neurons in BGstructures through electrodes implanted for direct brain stimulation (DBS)[211, 213, 214]. Since then, many studies have demonstrated that oscillatoryactivities of the BG are found in frequency bands ranging from low delta (2–4 Hz) to high-gamma (250–330 Hz). In particular, neural oscillations in thebeta band (13–30 Hz) appear to reflect motor states of PD patients [308]:PD patients in an off-medication condition have enhanced beta oscillations,and following administration of levodopa medication this beta power is de-creased in both the subthalamic nucleus (STN) and GPi [50]. The beta-bandLFPs have also been shown to correlate with movement preparation and ex-ecution as well as motor performance in PD patients [417]. One hypothesisfor the functional association of exacerbated beta oscillation with PD is thatnormal motor command for initiation of movement cannot override it, re-sulting in difficulty of generating voluntary movement for PD patients [155].Apart from the abnormality in the beta-band, neural oscillations in lowerfrequencies have also been suggested to relate to dopaminergic medicationresponses of PD patients. Oscillatory activities in the 4–10 Hz range havebeen shown to increase after dopaminergic medication and correlate with theimprovement in clinical condition [307, 350], and abnormal synchronized os-cillations around 8 Hz have been shown to correlate with levodopa-induceddyskinesia [9, 112, 351].Compared to the well-characterized oscillatory characteristics of PD pre-sented in the subcortical structures, how PD influences functional neuralnetworks in cortical regions (which can be relatively easily assessed with theelectroencephalography—EEG) are unknown. In early EEG studies, one ofthe findings presumed to be relevant to PD was slowing of neural rhythmsand resultant increased neural activity in low frequency bands (<10 Hz). Itwas postulated that occipital slowing may have resulted from the subcorticalstructures affected in PD [260]. However, a limitation of the argument isthat slowing of the occipital peak frequency is not specific to PD and it hasbeen commonly observed in people with other neurodegenerative conditions,such as Alzheimers disease (AD) [163, 285].Multimodal studies that record electrical potentials simultaneously fromthe cortex and the subcortical structures may provide useful clues in searchfor neurophysiological biomarkers specific to PD. For example, it was foundthat recordings of LFPs in GP and EEG in the supplementary motor ar-eas were closely related at <10 Hz and in 20–30 Hz when a PD patientwas off-medication [47]. As the pathological neural activities in the sub-cortical structures were represented in the cortical areas, it is natural to41.1. Parkinson’s Disease (PD)ask whether it would be possible to distinguish between Parkinsonian andnon-Parkinsonian states inferring from cortical activity alone. This couldprovide neurophysiological non-invasive PD biomarkers and would be es-pecially important in developing non-invasive brain stimulation techniquesaiming to provide therapeutic benefits to PD populations.1.1.4 Current Treatment Options for PDPD is normally treated pharmacologically with administration of dopamin-ergic medications, such as the dopamine precursor levodopa (L-dopa), thatis converted to dopamine after crossing the blood brain barrier. While earlyresponse to the medication is robust and satisfying, the medication likelydoes not alter progression of the disease. Prolonged use of the medication of-ten induces dyskinesia (involuntary hyperkinetic movement) and end-of-dosedeterioration (early wearing off) that can cause motor fluctuations betweenbeing “on” and mobile and “off” and stiff. In addition, not all symptoms areL-dopa responsive. L-dopa has little effect on gait and balance dysfunctionand non-motor symptoms such as autonomic dysfunction, sleep disorders,mood disturbances and dementia [345].DBS is a surgical treatment option for people with advanced PD. DBSelectrodes are implanted into the target structure in the brain to send electri-cal signals and a battery pack/implanted pulse generator (like a pacemaker)is inserted into the chest. Since the first use in 1986 of DBS with electrodesimplanted in the ventral intermedius nucleus of the thalamus to treat tremorin PD, DBS has been developed into an effective treatment for several medi-cally refractory movement disorders. In PD, DBS of the thalamus, GPi andSTN at high-frequency (>100 Hz) has been an effective and safe interventionand attenuate pathological neural oscillations. A five-year follow-up studyreported that 130-Hz DBS at the STN effectively treated motor symptoms ofpeople with advanced PD, resulting in general improvement in rest tremor,rigidity, gait, and akinesia as well as persistent improvement in motor symp-toms [188]. Dyskinesia was also found to be alleviated by 50–70% in otherstudies [89, 181].Present understanding of the therapeutic effects of DBS from functionalimaging, neurochemistry and neural recording studies suggests two stronglydebated general hypotheses: 1) DBS acts as a functional ablation to suppressthe stimulated nucleus, which is analogous to lesion of target structures inthe thalamus or BG, or 2) DBS results in activation of the stimulated nucleusthat is transmitted throughout the network [244].Although DBS can provide some benefits in managing motor symptoms51.2. Non-invasive Brain Stimulation (NIBS)in PD, it has several limitations. DBS is less effective for medication-unresponsive symptoms such as postural imbalance, freezing of gait andnon-motor symptoms [125, 318]. Furthermore, several studies reported cog-nitive decline (in particular executive function), reduction of verbal flu-ency, transient neuropsychiatric symptoms including hypomania, impulsecontrol disorders or hypersexuality, and suicidal ideation as side effects ofDBS [77, 131, 274, 282]. Surgery-related complications such as intracerebralhemorrhage and postoperative infections remain a possibility [131], whichcan be increased by periodic replacement (mostly every 4 years) of the bat-tery of the controllers and hardware malfunctions including lead breakageor malfunction of the pulse generator [225, 406].1.2 Non-invasive Brain Stimulation (NIBS)1.2.1 BackgroundWith a growing consensus on the important role of abnormal dynamics ofthe neural network involved in PD, non-invasive brain stimulation (NIBS)has been attracting substantial attention as a safe and effective means of dis-rupting abnormal oscillations. NIBS refers to stimulation techniques thatdo not require an incision or insertion in the body for electrode placement.The field is growing exponentially as NIBS methods are recognized as animportant tool to probe brain-behaviour relationships [300]. While the in-ferences from brain imaging methods alone are purely correlative, combinedwith NIBS to causally manipulate neural activity, the methods allow fordirectly studying how the altered neural activity causally affects behaviour.The most established NIBS techniques are transcranial magnetic stim-ulation (TMS) and transcranial electrical stimulation (tES). As they in-duce electrical fields over relatively large areas of tissue, the spatial focalityis much lower than invasive methods (Fig. 1.2) and generally entail neu-roimaging methods and computational modeling to visualize and interpretthe affected brain areas. Focused ultrasound stimulation (FUS) is a rela-tively new method and known to change neuronal activity with a resolutionof millimeters [386]. Although successful modulation of event-related poten-tials (ERP) in primary somatosensory cortex in humans was reported in arecent paper [208], safety studies and further research are required in thefuture to explore capabilities and limitations of the technique [388].61.2. Non-invasive Brain Stimulation (NIBS)Figure 1.2: The spatial and temporal resolution at which different brain interventionswork. NIBS methods work at the mesoscale level with the temporal resolution varies be-tween high and low depending on the specific type of the stimulation. Figure modified fromPolana and colleagues [300]. (tFUS: transcranial focused ultrasound stimulation; rTMS:repetitive transcranial magnetic stimulation; sTMS: single-pulse transcranial magneticstimulation; tDCS: transcranial direct current stimulation; tRNS: transcranial randomnoise stimulation; tACS: transcranial alternating current stimulation)1.2.2 Stimulation Parameters and ProtocolsThe basic principle underlying TMS is that time varying magnetic fields gen-erate electric fields. TMS applies strong but short (∼1 ms) magnetic pulsesto the scalp through a coil, inducing an electrical field in the brain and de-polarizing cell membranes [21]. The effects of TMS depend upon a numberof effects, including the geometry of the stimulating coil with respect to thehead, the frequency, intensity and pattern of the magnetic pulses [320], andthe duration of the stimulation. The spatial resolution of TMS for corti-cal stimulation is relatively higher (a few square centimeters in the cortex)than tES [223] and specific coil designs may also allow for stimulation ofdeep brain structures [87]. Single-pulse TMS delivers a monophasic pulseand evaluates excitability and conductivity of corticospinal motor pathways[186], and paired-pulse TMS consists of two successive pulses delivered withan inter-stimulus interval ranging from a few milliseconds to hundreds ofmilliseconds and allows the investigation of intracortical mechanisms of in-hibition and facilitation [320]. Repetitive TMS is a new generation of TMSintroduced in the late 1990 that delivers biphasic pulses repetitively with alow (< 1 Hz) or high (up to 60 Hz in general) frequency [319], and can leadto long-lasting after-effects compare to the single-pulse TMS [186].71.2. Non-invasive Brain Stimulation (NIBS)tES applies electrical currents to the brain through two or more stim-ulation electrodes attached to the scalp with conductive gel. tES methodsare categorized into transcranial direct current stimulation (tDCS), tran-scranial alternating current stimulation (tACS), and transcranial randomnoise stimulation (tRNS) depending on the stimulation waveforms. tDCSinduces constant depolarization and hyperpolarization to the cortical neu-rons close to anodal and cathodal electrodes that are fixed, whereas tACSapplies time-varying current (i.e., the anodal and cathodal electrodes are notfixed) with a single or multiple frequencies usually in a range of the oscilla-tory frequencies of the brain [17]. tRNS uses random values with particularprobability distributions as the stimulation current [322], which is currentlynot as common as tDCS and tACS.As the primary technique used in this dissertation is electrical stimula-tion, the following two sections (1.2.3 and 1.2.4) describe proposed mecha-nisms and clinical effects on PD with respect to transcranial electrical stim-ulation.1.2.3 Proposed mechanisms of tESPriori and colleagues [306] conducted the first modern study demonstratingmodulation of cortical excitability with tDCS, whereby anodal and catho-dal tDCS on the motor areas affects motor-evoked potential elicited in handmuscles in 15 subjects in an opposite direction. This was confirmed by thestudy of Nitsche and Paulus [264] that showed anodal tDCS augments mo-tor cortex excitability and cathodal DCS produces the opposite effect. Thispolarity-specific effects have become the reference for subsequent studies,which have demonstrated that the stimulation current can induce the re-sponse in the form of plasticity such as long-term potentiation (LTP) orlong-term depression (LTD) [284]. However, it is becoming more evidentthat the underlying processes involved in tDCS is more complex than sim-ply observing anodal and cathodal tDCS decreases and increases excitability,respectively, as the effects also depend on other stimulation parameters suchas stimulation time and intensity [24, 255]. Beyond the most accepted effectsof tDCS to change threshold for action potential generation by modulatingneuronal membrane polarity [217, 363], a number of cellular and molecularpathways are also affected by tDCS [284] with the mechanisms underlyingthese changes still being actively explored.For tACS, an important mechanism is entrainment of oscillatory brainactivity at or near the stimulation frequency. Zaehle and colleagues [436]demonstrated that 10-min tACS applied at individual’s EEG alpha fre-81.2. Non-invasive Brain Stimulation (NIBS)quency (IAF) increased the post-stimulation EEG spectral power specificallyin the range of the IAF, indicating that tACS can induce frequency-specificeffects on brain oscillations measured by EEG. This was replicated in a fol-lowing study [261] and additionally it was reported that the increased alphapower persisted for at least 30 min after stimulation cessation. Entrainmentis a theoretical concept originally conceived to explain synchronization phe-nomena in nonlinear systems [293] and described by the so-called “Arnoldstongue” that predicts the degree of synchronization of an oscillator with agiven natural frequency to a rhythmic driving force as a function of drivingforce amplitude and frequency [267]. In the context of neural oscillations, thefollowing list of features formulated by Thut and colleagues [382] can guideone to determine if brain oscillations induced by tACS qualify as “neuralentrainment”:1. Entrainment requires the involvement of a neural oscillator (i.e., a neu-ral population that exhibits oscillations at the entrainment frequencyor is capable of doing so under natural conditions).2. Entrainment requires periodicity in the input stream of external events.The external events can be in any form (e.g., electric, magnetic, vi-sual) and have any periodic shapes (e.g., sinusoidal, a square-wave,repeated pulses).3. Entrainment requires synchronization (phase alignment) between theinput stream and the neural oscillator.4. Crucially, the models also assume that the external force influencesthe oscillating elements by direct interaction (i.e., there should be nosecondary stages such as connected brain areas).Although this framework provides a conceptual framework for investigat-ing the mechanisms of tACS, this model is clearly a coarse approximation tothe mechanisms governing the brain and does not explain phenomena suchas the fact that entrained brain oscillations are not always found after tACS[408] and it can vary depending on other parameters such as stimulationintensity and duration [409]. Another line of reasoning for tACS effects atlarge-scale network level has been recently proposed. This network activity-dependent model is based on the fact that our brain is a network consistingof spatially distributed but functionally linked regions [397] and electricalstimulation induces an activity-dependent modification of the system notonly in a local area but also in specific networks [33, 224, 250]. In this91.2. Non-invasive Brain Stimulation (NIBS)approach, tACS effects on neuronal activity and behaviour outcomes aredependent on the on-going state of relevant brain networks [106] that maycooperate or compete with each other. In support of this, several studieshave demonstrated that behaviour changes via same type of stimulation canvary depending on the level of network engagement induced by the task[106].The mechanisms underlying tRNS are largely unknown, but stochasticresonance is considered as one potential mechanism. Stochastic resonance(or stochastic facilitation in some fields) describes the contribution of addednoise to a nonlinear dynamic system, and according to this view, the in-jection of appropriate level of random noise can paradoxically enhance theresponse of a nonlinear system (e.g., nervous system) to a weak signal [237]depending on the intensity of the noise and the state of the system. For in-stance, it was demonstrated that the ability to detect a subthrehold tactilestimulus in healthy subjects was enhanced when receiving the subthresholdstimulus with particular level of noise compared to receiving the stimulusalone [71]. As for tRNS, this framework views the week-current applied viastimulation as the introduced noise and activations of a network of neu-rons responsible for executing a specific process or function as the state ofthe signal, and their interaction may in part can explain several cases offacilitatory or inhibitory effects of stimulation [106, 296, 398].1.2.4 Clinical Research Findings on PD using tESDespite the field of tES has expanded rapidly over the last decades, mosttherapeutic studies in PD have applied rTMS, and tES remains a prospectivetherapeutic tool [28]. A summary of literature reviews on tES studies in PDis provided in Table 1.1. Ten tDCS studies reported therapeutic effectson motor and cognitive functions in PD and three tACS studies reportedmodulatory effects on cortico-muscular couplings. No tRNS study was foundwith respect to PD.1.2.5 Challenges and Open QuestionsTo understand the neurophysiological mechanisms and ongoing effects ofNIBS, it is necessary to monitor the changes in the brain activity using neu-roimaging techniques. EEG and magnetoencephalography (MEG) are twomost widely used neuroimaging techniques with NIBS, and to date electro-physiological changes in the brain have mostly been investigated by compar-ing the recordings before and after stimulation due to the strong artifact that101.2. Non-invasive Brain Stimulation (NIBS)Table 1.1: A summary of tES research on PDType 1Methods CurrentIntensityDuration 2N Outcome Year(Ref.)DC A: premotorC: mastoids2 mA 20 min (8sessionswithin 2.5weeks)25 PD - Improved gait and bradykinesia- No effect on UPDRS, reaction time,physical and mental wellbeing,self-assessed mobility2010 [29]DC A: motorC:orbitofrontal2 mA 20 min (5consecutivedays)10 PD - Improved total and motor UPDRSscores, FOG-Q and Gait and FallsQuestionnaire scores2014[395]DC A: DLPFCC:supraorbital2 mA 20 min 18 PD - Improved accuracy in working memorytask2006 [38]DC A: DLPFCC:supraorbital2 mA 20 min 16 PD - Enhanced verbal fluency- Improved phonemic fluency task2013[287]DC A: DLPFCC:supraorbital2 mA 20 min (10sessions over2 weeks)18 PD - Improved executive function 2014 [92]DC A: DLPFCC:supraorbital2 mA 7 min 10 PD - Improved locomotor performance 2014[230]DC A: M1C:supraorbital1 mA 20 min 17 PD - Improved UPDRS, simple reactiontime, motor-evoked potential2006[114]DC A: right M1C: left M12 mA 25 min 10 PD15 H- Decreased noise in arm movement- Increased willingness to exert effort2015[329]DC A: Cerebellaror M1C: deltoidmuscle2 mA 20 min (5consecutivedays)9 PD - Decreased UPDRS IV (dyskinesiassection) score2016[105]AC 20 Hz at M1 1 mA 15 min 10 PD10 H- Decreased cortico-muscular coupling in13–30 Hz2014[189]AC 77.5 Hz atfrontal area15 mA 45 min (10sessions over2 weeks)23 PD - Insignificant changes in UPDRScompared with sham- Significant changes in UPDRS frombaseline2011[348]AC Individualtremorfrequency atcerebellum2 mA up to 10 min 24 PD21 ET- Intrained tremor phase to stimulation 2015 [45]AC Individualtremorfrequency atM12 mA 10 min 14 PD - Average 50% reduction in tremoramplitude2013[151]1A and C denote anode and cathode, respectively2PD: Parkinson’s disease, H: healthy, ET: essential tremor111.2. Non-invasive Brain Stimulation (NIBS)NIBS produces in both EEG and MEG recordings during the stimulation[151, 249, 381]. In fact, these stimulation artifacts can be up to 3 orders ofmagnitude larger than normal brain signal, completely obscuring EEG andMEG signals. In addition, the delivered current can undergo possibly non-stationary transformation, causing the morphology of stimulation artifactsin EEG and MEG recordings do not exactly match the delivered currentsat the stimulation electrodes, making it even more technically challengingto remove the artifacts. For example, tACS stimulation with a single sinu-soidal wave generates EEG and MEG artifacts with the same frequency buttime-varying amplitudes and shifted phases [268, 269]. A thorough charac-terization of the stimulation artifacts and development of effective denoisingmethods are required to address these issues.One crucial unresolved issue is the question as to whether tES protocolselicit their strongest effects under the electrodes. Due to the high conduc-tivity of the skin compared to the skull, most of the stimulation current runsthrough the skin and only a small fraction of the applied current actuallyreaches the brain. This leads to not only a decrease in the effectivenessof the stimulation (as one can only endure so much high intensity of thecurrent) but also focality of the stimulation. This poor spatial resolutioncan be improved at the expense of intensity by optimizing the arrangementof stimulating electrodes using details of a subjects anatomy and computerstimulations [91, 323], which is currently an active area of research.The effects of stimulation are rather heterogeneous and there is a lackof replicability across studies, raising concerns with regard to the validityand reproducibility of the results [151]. This is partially attributed to largeoptions for stimulation parameter selection including stimulation frequency,intensity and electrode montage, leading to a lack of consistency in stimu-lation protocols adopted in each study. Moreover, a range of cognitive andbehaviour tasks used and study populations add to the variation of stim-ulation outcomes. Other part of this heterogeneity may be explained bypoor focality of the NIBS and intersubject variability at baseline that arenot accounted in the study. Therefore, continuous efforts such as providingrationales for selected stimulation protocols, conducting replication studies,and identifying factors for heterogeneous results are recommended for futureNIBS studies.Although remarkable progress in the understanding of NIBS has beenmade, mechanisms that could explain the stimulation effects still remainlargely unknown and each model proposed to date provides conceptualiza-tion and explanation of the observed phenomenon from a slightly differentpoint of view [106]. Underlying mechanisms are likely explained when inte-121.3. Electrical Vestibular Stimulation (EVS)grating the different concepts rather than a single explanation. New theo-retical models are currently being actively developed based on experimentalevidence and computational simulations.1.3 Electrical Vestibular Stimulation (EVS)1.3.1 BackgroundElectrical Vestibular Stimulation (EVS) is a NIBS technique where electricalcurrent is applied to the mastoid process behind the ear to alter the firingrates of the vestibular afferents. Stimulation of vestibular nerves by EVSultimately influences the activity in various cortical and subcortical areasrelated to the vestibular network and multisensory processing including theprefrontal cortex, premotor region, somatosensory cortex, posterior pari-etal cortex, intraparietal sulcus, inferior parietal lobule, temporo-parietaljunction, insula, hippocampus, and putamen. While the mechanisms arenot fully established, it has been proposed that EVS activates these regionsbased on the broadly distributed thalamocortical fibres distributed through-out numerous brain regions [221].EVS has many advantages as an investigative and potential therapeu-tic approach. Different from tDCS/tACS where it is currently difficult toknow which brain regions the externally applied electrical current is deliv-ered [325], EVS bypasses the vestibular end organ and acts directly at thespike trigger zone of the afferent nerve [111, 128]. Because EVS allows forthe delivery of precise levels of applied electrical current, it is well-suited forsubliminal stimulation so that the subject is unaware that they are receivingverum or sham stimulation, often precluding the additional requirements oftrials to control for the placebo effect in clinical studies. Furthermore, thelow currents typically involved suggest the feasibility of battery-powered,portable stimulation. In comparison to other means of vestibular stimu-lation, such as caloric, EVS does not commonly induce adverse side effectssuch as seizures, vertigo or nausea; however, symptoms of tingling and slightitching underneath the electrodes have been reported [393].The results of EVS are varied and complex, reflecting the complex roleof the vestibular system. While the primary induced physiological responsesfrom EVS are gaze stabilization, posture and balance maintenance and self-motion perception, this requires integration of visual, proprioceptive andsomatosensory information from the earliest thalamic stage to cortical net-work interactions [75]. As a result, the vestibular system has a broad range offunctions from reflexes (e.g., vestibulo-ocular and vestibulospinal) to higher131.3. Electrical Vestibular Stimulation (EVS)levels of voluntary motor and even cognitive behaviour such as visual mem-ory recall [422, 424] and mental object transformation strategies [209]. Thus,augmenting vestibular input has been investigated for number of conditions,including neuropathic pain [240, 241], tactile extinction [339], and figurecopying deficits [423].1.3.2 The Vestibular SystemThe vestibular system, consisting of three semicircular canals, saccule, andutricles in the inner ear, is a sensory system that provides the sense ofmotion, equilibrium, and spatial orientation to the brain. It is different fromother senses in that central vestibular processing is highly convergent andmultimodal as signals from muscles, joints, skin, and eyes are continuouslyintegrated with vestibular inflow [16]. Because of the features, the vestibularstimulation does not induce a separate and distinct conscious sensation andcontributes to a range of functions from simple reflexes to the highest levelsof perception and consciousness [16].Anatomically, the vestibular nerve combines with the cochlear nerve andbecomes the vestibulocochlear nerve. Traveling by the cerebellopontine an-gle, this nerve enters the brainstem at the pontomedullary junction in whichthe vestibular and cochlear nerves are separated out [173]. Some of the nervefibers project to the flocculonodular lobe and nearby vermis of the cerebel-lum while the majority of the fibers projects to the ipsilateral vestibularcomplex in the pons [173]. The vestibular complex is where vestibular in-puts are primarily processed and consists of four major nuclei includingmedial, superior lateral and inferior [421] and several adjacent cell groups.The vestibular pathways from the vestibular nuclei can be functionally cat-egorized. Projections to the spinal cords are essential for postural reflexesto adjust the head and body movement [167], and projections to the ocu-lar motor nuclei are critical for compensatory eye movements during headmotion (i.e., vestibular-ocular reflex). Projections to the cerebellum are im-portant for balance, postural control, and movement coordination [173], andthe pathways to the thalamus, hippocampus and ultimately to the corticalareas are responsible for multisensory integration [368, 421] contributing tomovement planning and execution, spatial navigation and memory, atten-tion, and emotions [43, 134, 312, 314, 368].141.3. Electrical Vestibular Stimulation (EVS)1.3.3 EVS Effects on PDSeveral studies have investigated effects of EVS on postural responses in PD.Compared to a group of age-matched control subjects, PD patients showedno difference in the speed or direction of the body sway response inducedby EVS (2 s), but when the patients were subdivided into two groups, thepatients with greater postural deficit responded with significantly greaterbody speed than those with milder postural deficit [283]. In another study,EVS was applied for a longer duration of 20 minutes to PD patients withpostural instability and it was found that the instability was reduced afterthe EVS [171]. Similarly, EVS improved body sway in the anteroposteriorand mediolateral directions [277], balance corrections and postural responsetime after a backward perturbation [330], and anterior bending posture inPD patients [272], suggesting that EVS may be beneficial to balance andposture control in PD. In support of this, a recent rodent study demonstratedthat EVS improved balance and motor planning of a 6-hydroxydopaminehemilesioned rat model in the accelerating rod test [331].EVS has also shown beneficial effects on motor symptoms in PD. Ya-mamoto and colleagues [432] demonstrated that EVS ameliorated autonomicand motoric disturbances and decreased reaction time in Go/NoGo taskswithout affecting omission and commission errors. Akinetic symptoms ofPD patients were improved after 24-hour EVS [278], and EVS improvedmotor symptoms of upper and lower extremities as measured in finger tap-ping task and the Instrumented Timed Up and Go test [174]. Recent ev-idence that showed EVS induces significant neurochemical changes in thestriatum may partially explain mechanisms underlying the motoric effectsof EVS [369]. In summary, EVS may possibly carry an effective therapeuticbenefit to improve autonomic and voluntary motor responses for PD pa-tients, although rigorous assessments with quantitative motor metrics anddemonstrated efficacy beyond that provided by medication are needed.1.3.4 Stimulation ParametersVarious stimulation parameters including signal types (i.e., DC, AC, or noisy(stochastic)), frequencies, and current intensities have been utilized in EVSstudies to elicit neurological or physiological responses of interest (Table1.2). A large number of EVS studies on PD have been based on noisy stim-ulation (i.e., randomly varying stimulation currents) [278, 331, 424, 432].In particular, rather than broadband white noise, pink noise with a 1/ftype power spectrum (i.e., the power density of the stimulus is inversely151.3. Electrical Vestibular Stimulation (EVS)proportional to the frequency) is used, as this reflects the power distribu-tion found in cortical and subcortical functional networks [56]. Similar totRNS, one of the justifications to explain how the randomly-varying stimulimay provide beneficial effects is a stochastic resonance phenomenon wherea sub-threshold random stimulus enhances sensory information processingand perception [257]. For example, 40 Hz responses of the human auditorycortex to auditory stimuli was enhanced when weak noise was added to thestimulus [413]. However, it was recently shown that induced EEG changesfrom noisy EVS tend to be linearly related to the intensity of the noise levelin healthy subjects [177], which may appear inconsistent with stochasticfacilitation seen in non-linear systems. A reasonable explanation for thisapparent discrepancy is that while stochastic facilitation may be apparentat the level of the individual neuron, collectively at the overall network level,linear responses may prevail although this needs to be further verified.Taken together, it is clear that comprehensive investigation needs to bedone to understand the underlying mechanisms of EVS effects and ultimatelymaximize clinical effects it can bring. Especially, one crucial component tostudy would be to design stimuli with properly selected parameters to ex-tract sufficient information of neural responses rather than blindly choosingstimulation parameters as has been done in previous work. As there areliterally an infinite number of ways to select stimulation parameters, morerigorous and systematic approaches are warranted.1.3.5 Multisine SignalThe majority of EVS studies have used general-purpose stimulus such assquare-wave pulses and random (white or pink) noise to investigate stimula-tion effects on vestibular function and corresponding physiological responses.A significant limitation of the stimulation method is relatively small re-sponses induced by these stimuli [81, 110], possibly due to poor excitationof the neurological system. Although greater responses might be induced byhigher stimulus amplitudes, the stimulus level is often restricted by the rangewhere subjects feel comfort or nature of the study where subjects need tobe blinded from stimulation to avoid the placebo effect. Another limitationof general-purpose signals is that they do not provide enough information ofthe systems under study and are difficult to set signal parameters to achievean optimal result.Multisine signals are optimized test signals utilized most commonly inthe field of system identification, which are designed to concentrate power ata precise number of frequencies within the bandwidth of interest. They are161.3. Electrical Vestibular Stimulation (EVS)Table 1.2: Effects of varying EVS parameters on physiological responsesType CurrentIntensityDuration 1N Outcome Year(Ref.)DC 0.3, 0.5 mA 8 s 10 H Increased postural sway with highercurrent2003[414]DC 0.1-0.9 mA 2 s 10 H Tilted head and torso toward anodalelectrode1997 [80]DC 0.7 mA 3-6 s 12 H Tilt of the body dependent on the timingof stimulation with respect to movementphases1998 [62]DC 1.5-3 mA 5 s 12 H Ocular torsion and rotation of the foveaand peripheral visual field1997[438]DC 0.1-0.9 mA 4 s 6 H Ocular torsion and horizontal eyemovements at higher currents2003[346]DC 2 mA 20 s 6 H Torsional eye movements 2002[340]DC 1.25, 2.5 mA 20 s 14 H Tilt of the subjective vertical 2001[232]DC 0.8-1.2 mA 15 s 8 H Increased response time for bodyrotation and illusory sensation of motionof self or visual field2001[209]DC 0.2, 0.5, 0.7mA20 min 7 PD Improved anterior bending posture 2015[272]DC 0.7 mA 20 min 5 (5) PD Reduced postural instability for 3 out of5 patients with PD with posturalinstability and/or abnormal axial posture2001[171]DC Twice of theindividualthresholdDuring task 11 PD Improved variation of the step durationin gait and improved motor performancein finger tapping task2018[174]AC (0.2,0.5, ..., 2Hz)2 mA 200 cycles (1min 40 s - 16min 30 s)14 H Postural illusions of ‘rocking’ or‘swinging’ and vestibular modulation ofmuscle sympathetic nerve activity2018[174]AC (0.2,0.5, ..., 2Hz)2 mA 200 cycles (1min 40 s - 16min 30 s)11 H Postural illusions of ‘rocking’ or‘swinging’ and increased burst incidenceof skin sympathetic nerve activity2010[159]AC (1 Hz)1/f noise(0.1-10 Hz)90% ofindividualthreshold5 min 23 PD / 12 H Increased overall connectivity ofPedunculopontine Nucleus with 10regions of interest2018 [57]Noise(0-1000 Hz)90% ofindividualthresholdDuring task 24 H Shorter reaction time with thestimulation when answering questionsabout faces2008[424]171.3. Electrical Vestibular Stimulation (EVS)Type CurrentIntensityDuration 1N Outcome Year(Ref.)Noise(frequencyis notreported)0.5-1.5 mA 25 min (1-5sessions)49 H Improved hemispatial neglect 2004[425]1/f Noise(0.1-10 Hz)90% ofindividualthreshold72 s 10 H A mild suppression of gamma power inlateral regions / Increased beta andgamma power in frontal regions2013[177]1/f Noise(0.01-2 Hz)60% ofindividualthreshold24 h 12 H / 7MSAImproved autonomic system regulation /Decreased reaction time to visual cue2005[432]1/f Noise(0.01-2 Hz)0.09-0.49 mA 24 h 8 (8) MSA /3 PD / 2CCA / 1 PAImproved motor function in patients withPD2008[278]1/f(frequencyis notreported)0.1, 0.3, 0.5mA26 s 5 PD / 20 H Decreased body sway with eyes closed 2009[277]White noise(0-30 Hz)Belowindividualthreshold< 3 h 10 PD Improved balance corrections after abackward perturbation, shorted thepostural response time2015[330]White noise(frequencyis notreported)70 % ofindividualthreshold60 s 13 PD / 12 H Changes in posture and increased swayamplitude and mildly decreased swayfrequency2018[384]1PD: Parkinson’s disease, H: Healthy, MSA: multi system atrophy, CCA: cortical cerebellar atrophy, PA: pureakinesiaadvantageous to target specific components of the responses, enable consid-erable reduction of the measurement time without unwanted loss of accuracy,and can be used to detect and quantify the presence of nonlinear distortionsin the system. With the properties, the multisine signals are considered tobe advanced test signals than general-purpose excitation signals such as theswept sine (also called periodic chirp) or random noise that are applied tothe system without any optimization aside from selecting the bandwidthof the excitation signal [294]. Multisine signals are composed of sinusoidswith period equal to (or integer ratios of) the observation time, which keepsexcitation power as low as possible outside frequencies of interest avoidingunnecessary nonlinear effects [353]. Further optimization is done by choos-ing the frequency phases such that the crest factor (CF), defined below, of181.3. Electrical Vestibular Stimulation (EVS)the signal to be minimized [294, 341, 342]:CF =max |Istim(t)|RMSIstim(1.1)where RMSI is the RMS of the applied current Istim(t).It is advantageous to excite the system with the optimized multisinesas the signals with a large CF inject much less power into the system thanthose having the same peak value and a small CF [294].Multisine Signals for EVSMultisine signals for EVS were designed through two steps, the first beingthe selection of a period (i.e., frequency resolution) as well as a frequencyband of interest followed by a crest factor minimization of the signals inthe second step. The 4–200 Hz frequency band was chosen as it includesthe range human EEG responses [150]. The CF of all multisine EVS wasminimized using a clipping algorithm, an iterative method developed in[400, 401] to optimize the phases. The basic idea behind this method isillustrated in Fig. 1.3. With the specified amplitude spectrum, the iterationprocedure starts with arbitrary phases and a discrete time-domain signal iscalculated by the inverse Fourier transform. All the values larger than agiven maximum is clipped off to generate a new time signal whose spectrumand phases are calculated using the FFT. These new phases are retained asa first approximation to the solution [294], but the amplitude spectrum isrejected in favor of the original one.In this dissertation, all multisine signals were designed to have a periodof 5 s [113], providing a frequency resolution of 0.2 Hz. Seven kinds ofmultisine stimuli were designed over the frequency range from 4 to 200 Hz(Table 1.3). The division of the frequency bands between 4–50 Hz refers tocanonical EEG frequency bands.Table 1.3: Seven multisine stimuli and corresponding frequency bandwidthsMultisine Stimulus 1 2 3 4 5 6 7Frequency (Hz) 4–8 8–13 13–30 30–50 50–100 100–150 150–200191.4. Research Objectives and Thesis OutlineFigure 1.3: Clipping algorithm for minimizing crest factor of multisine signals [294]1.4 Research Objectives and Thesis Outline1.4.1 Research objectivesThe therapeutic potential of EVS has been demonstrated in previous studies– even with applying relatively simple stimuli such as DC or random noise– by showing voluntary motor and postural responses in people with PD.However, despite many years of the research, there is a huge lack of 1)understanding of the relations between stimulation parameters and resultantbehaviour responses in PD and 2) brain imaging studies that can provideinsight into underlying neurological mechanism of EVS effects.The goal of this dissertation is to advance application of EVS as a poten-tial therapeutic intervention for PD through development of novel stimuliand thorough investigation of neuronal and behaviour effects by utilizing be-haviour tasks and brain imaging modalities including EEG and functionalMRI (fMRI). This research will particularly investigate subthreshold EVSeffects so that it could provide the foundation for a safe, non-invasive, andultimately portable ancillary therapy for PD patients, focusing on the fol-lowing objectives:• Objective 1. Development of new EVS stimuli and design ofexperiment: The first objective is to design new stimulus candidatesthat can systematically provide information on the effects of stimu-lation parameters on brain activities. As mentioned above, multisinesignals have advantages as exogenous input signals to perturb a sys-tem (i.e., the brain) over random noise and single-frequency sinusoids.201.4. Research Objectives and Thesis OutlineMultisine signals in distinct frequency bands are proposed as stimu-lus candidates (see 1.3.5 for details). As a non-invasive brain imagingmodality, EEG has merits that it can measure the brains electricalactivity with high temporal resolution (a millisecond) and is relativelyinexpensive compared with other technologies and is simple to oper-ate. Thus, EEG is proposed as a primary modality to monitor brainactivities before, during and after EVS. Experiments are designed toinvestigate brain responses to EVS 1) when subjects are resting and 2)when subjects are performing a motor task. Motor task performance ofthe subjects is used as behavioural outcomes from the EVS. Finally,fMRI is utilized as a primary brain imaging modality to investigatemechanism of EVS as it can measure brain activities (hemodynamicresponses) with excellent spatial resolution.• Objective 2. Detection of PD-related features in the EEGand fMRI data: A two-step approach is proposed to evaluate EVSeffects in PD patients. The first step is to identify PD-related featuresin the EEG and fMRI data collected in the pre-EVS condition (i.e., be-fore applying EVS). To do this, it is proposed to collect the data fromPD and healthy subjects and extract features to discriminate brainactivity of the two groups (e.g., functional connectivity). For the PDgroup, the data are collected in off- and on-medication conditions inorder to compare effects of each of the EVS and medication interven-tions and their interactions in PD. Following successful identificationof the PD-related features, the second step is to assess modulatoryeffects of EVS on the features (Objective 3).• Objective 3. Establishment of the extent that EVS modu-lates the PD features and improves motor performance: Theeffects of EVS on the identified PD features are investigated focusingon addressing the following questions:– Is the stimulation able to normalize PD features?– How do the stimulation effects vary according to stimulation pa-rameters?– Are there any differences between online- and after- effects of thesimulation?– Are there group-specific effects (i.e., PD vs. healthy controls)?– Is the stimulation able to induce significant improvement in motorbehaviour of the PD patients? How do the downstream behaviour211.4. Research Objectives and Thesis Outlineoutcomes relate to stimulation-induced changes in the brain?• Objective 4. Development of a new method to remove stim-ulation artifacts in EEG: High-voltage electrical artifacts in EEGgenerated by brain stimulation have been a major challenge to date foranalyzing online effects of stimulation (during stimulation) on brain ac-tivities, which is critical to probe fundamental mechanisms underlyingstimulation effects. Most studies have resorted to avoiding the artifactproblem by simply comparing the EEG in pre- and post-stimulationcondition. A solution is proposed to resolve the artifact issue by devel-oping a novel denoising method utilizing joint blind source separationmethods.• Objective 5. Probing fundamental mechanisms of action throughwhich EVS improves motor performance: Noisy EVS has shownits efficacy to improve motor symptoms and postural responses in PDin prior studies, but the mechanisms of these effects are largely un-known. To probe these fundamental mechanisms, it is proposed toanalyze resting-state fMRI data acquired during noisy EVS from PDpatients and healthy controls, focusing on the thalamus, a hub of in-tegrating multisensory information and mediating functional networks[156]. The thalamus is of interest in particular based on the direct af-ferent projections from the vestibular nuclei and it’s close relationshipswith the BG.1.4.2 Thesis OutlineThe rest of this dissertation is subdivided into six chapters as outlined below:Chapter 2: Discriminant Feature Detection in Manual TrackingBehaviours in PD and Effects of Noisy EVSIn this chapter, effects of noisy EVS (0.1–10 Hz) on the manual trackingbehaviours of PD and healthy subjects who performed a visuomotor joysticktracking task. Exploratory (linear discriminant analysis with bootstrapping)and confirmatory (robust multivariate linear regression) methods are usedto determine if the presence of EVS significantly affected prediction of cur-sor position based on target variables, and signal-to-noise ratio of cursortrajectories is computed to quantify smoothness of tracking movement. Theresults show that noisy EVS resulted in robust changes in tracking, mostlyrelated to increased sensitivity to perceived error.221.4. Research Objectives and Thesis OutlineChapter 3: Quadrature Regression and IVA Approach to Removalof High-voltage EVS Artifacts from Simultaneous EEG RecordingsThis chapter describes the technical difficulties associated with remov-ing artifacts in EEG that are induced by electrical brain stimulation andlimitations of conventional denoising methods. Quadrature regression andsubsequent independent vector analysis (q-IVA) method is proposed for re-moving the stimulation artifacts and applied to simulated and real EEGdatasets recorded from ten subjects who received 4–8 Hz multisine EVS. Itis demonstrated that q-IVA significantly improves the denoising and robustlyrecovers the EEG compared to conventional methods (principal componentanalysis, independent component analysis) and other joint blind source sep-aration approach (multiset canonical correlation analysis and independentvector analysis). The results provide a promising way to effectively isolatesimulation artifacts in EEG, paving the way for future studies attempting touncover ongoing modulation of brain activity during electrical brain stimu-lation.Chapter 4: Sparse Discriminant Analysis for Detection of Patho-logical Dynamic Features of Cortical Phase Synchronizations inPDIn this chapter, altered cortical functional coupling in PD is identifiedusing resting-state EEG data and effects of multisine EVS at 4–8 Hz, 50–100Hz, and 100–150 Hz are examined. Phase locking value (PLV), a nonlin-ear measure of pairwise functional connectivity between electrodes, is com-puted over sliding windows and the mean, variability and sample entropyare extracted as dynamical features of the functional connectivity. To ex-tract most discriminant features from the high-dimensional data sets, sparsediscriminant analysis is utilized. It is demonstrated that lower PLV vari-ability and entropy in PD compared to healthy controls is normalized byEVS in a stimulus-dependent manner, suggesting that EVS with optimizedparameters may provide a new non-invasive means for neuromodulation offunctional brain networks.Chapter 5: Discriminant Correlation Approach to Joint Estima-tion of Maximal EVS Effects on Motor Behaviour and CorticalBeta Oscillations in PDUsing EVS and simultaneously recorded EEG, this chapter demonstratesthe modulatory effects of high-frequency (50–100 Hz and 100–150 Hz, respec-tively) multisine EVS on movement-related beta desynchronizations (beta231.4. Research Objectives and Thesis OutlineERD) and resultant changes in the motor behaviour of PD and healthy sub-jects who performed a motor squeeze task. In order to investigate maximalEVS effects across the subjects with regard to the task performance andthe beta ERD, discriminant correlation analysis, a feature fusion method,is used. It is demonstrated that EVS modulates the magnitude and timingof beta ERD in left motor, broad frontal and medial parietal regions duringperformance of the motor task. The beta power in the rest period, when thesubjects were not engaged in the motor task, was not significantly affectedby EVS. This joint EEG/behavioural analysis suggests a potential neuro-physiological mechanism of EVS in motor improvement, whereby vestibularinput is integrated in the motor thalamus, increasing fluidity to a motorsystem stuck in a state of exaggerated beta rhythms. The results com-plement previous studies suggesting pathological beta-band oscillations inPD can be disrupted via different stimulation sites, including ones availablenon-invasively, and emphasize the importance of stimulation parameters forinfluencing motor behaviour.Chapter 6: Spectral Clustering and Discriminant Correlation Ap-proach to Estimation of EVS Effects on Functional Thalamic Sub-regions and BG–thalamic Connectivity in PD—fMRI StudyPathologic changes within the thalamus itself and its functional interac-tions with the BG leads to altered cortical-BG-thalamo activity responsiblefor motor and cognitive dysfunction in PD. As the thalamus receives directprojections from the vestibular nuclei, it may possible to modulate thala-mic activity and connectivity with the BG by activating vestibular systemafferents with EVS. This chapter probes EVS effects on the thalamus usingresting-state fMRI data acquired from PD and healthy subjects to elucidatea potential mechanism of EVS associated with motor improvements in PD.To determine the region-specific EVS effects on the thalamus, normalizedcut spectral clustering is used to parcellate the thalamus into subregions anddiscriminant correlation analysis is applied to investigate functional connec-tivity between the thalamus subregions and BG structures. The resultsshow that EVS normalizes altered sizes of the functional thalamic subre-gions, reduces excessive connectivity between the right thalamic subregions,and improves aberrant asymmetry of the connectivity between left BG andbilateral thalami in the PD subjects.Chapter 7: Conclusion and Future Work241.4. Research Objectives and Thesis OutlineThis chapter includes a short of summary of the dissertation followed bya discussion on the limitations of the proposed methods and suggestions forfuture work.25Chapter 2Discriminant FeatureDetection in ManualTracking Behaviours in PDand Effects of Noisy EVSIn this chapter, we investigated effects of noisy EVS (0.1–10 Hz) on themanual tracking behaviour of PD and healthy subjects. Noisy EVS hasbeen recently used in prior PD studies to assess effects on motor symptomsand has positive influences on balance and simple motor task performance.Here, we implemented a visuomotor joystick tracking task to assess effects ofnoisy EVS on more complicated motor behaviours that require sensorimotorprocessing and fine motor coordination.2.1 IntroductionMotor symptoms in PD characteristically manifest themselves as tremor,rigidity, akinesia/bradykinesia and postural instability. While levodopa isthe gold standard treatment for PD, chronic use eventually leads to thelong-term development of side effects, such as motor fluctuations, dyskine-sias and psychiatric disorders [302, 418]. Surgical treatments, including DBStargeted to subcortical nuclei, have provided effective therapeutic benefits,but are complex and invasive [273]. With recent technological advances, nu-merous novel stimulatory techniques for PD treatment are presently beingexplored [101, 118, 331, 379]. Non-invasive brain stimulation techniques arecurrently a growing avenue of interest for PD and other neurological dis-orders due to their safety, tolerability and minimally invasive nature [115].Additionally, these methods, such as transcranial current brain stimulation(tCS), arguably influence solely the targeted site of stimulation, but alsoexert effects on associated brain connectivity patterns [224]. Since PD ischaracterized by abnormally exaggerated beta synchronization throughout262.1. Introductiona BG-cortical network [100], non-invasive stimulatory approaches could po-tentially be used to modulate aberrant network dynamics [115].A few studies have suggested that non-invasive stimulation of vestibularnerves via noisy EVS may improve motor deficits in PD [277, 278, 331, 432].Noisy EVS delivers currents with randomly varying amplitudes in time tovestibular afferents and subsequently influences resting state cortical EEGactivity, suggesting that cortical-subcortical connections are also modulatedby EVS [177]. Akin to how tCS strengthens connectivity patterns in premo-tor, motor and sensorimotor areas while subjects are engaged in a finger tap-ping task [299], noisy EVS hypothetically is also able to influence functionalBG-cortical motor networks depending on the brain state during stimula-tion. It is not fully established, however, whether noisy EVS improves motorperformance. Yamamoto et al. [432] measured trunk dynamics as well asreaction time in a Go/NoGo paradigm whereas Pan et al. [278] measuredwrist activity in akinetic PD patients. Effects of noisy EVS on posturaland balance responses have also been measured in both humans and ratmodels [277, 331], although none of these studies have directly investigatedthe effects of EVS on bradykinesia with respect to motor coordination andsensorimotor processing.One potential way to rigorously assess the motoric effect of EVS is toutilize a visuomotor task, which is useful for understanding mechanisms thatcontribute to motor coordination with accuracy and stability [324]. Correc-tive movements and behavior are required in response to varying visual errorfeedback, which are important for maintaining effective perception-action orsensorimotor processing [324]. With respect to clinical significance, the abil-ity to continually adapt ones behavior to changing environmental or sensorystimuli is particularly relevant in PD as these patients demonstrate impairedswitching between motor paradigms [97].In the present study, we implemented a visuomotor tracking task andinvestigated the effect of noisy EVS on motor performance. Our visuomotortask required subjects to respond to visual error feedback that was, unbe-knownst to the subjects, either minimized to 30% of the actual error, oramplified by 200% to create the appearance of Better or Worse motor per-formance, respectively. We used linear discriminant analysis (LDA) [96] toidentify parameters significantly influenced by EVS and to investigate if theeffects of EVS are dependent on the task conditions. We then analyzed ourdata using a robust multivariate linear regression method [108] to test iftracking movement was affected by EVS. We show that subthreshold EVSresulted in robust changes in tracking, mostly related to increased sensitivityto perceived error.272.2. Materials and Methods2.2 Materials and Methods2.2.1 Subjects12 PD subjects (10 males, 2 females; mean age 61.4 ± 6.5 years; 11 right-handed, 1 left-handed) participated in the study. None of the participantshad any reported vestibular or auditory disorders. All PD subjects wererecruited from the Pacific Parkinsons Research Centre (Vancouver, Canada).PD subjects had mild to moderate disease severity (Hoehn & Yahr stages1.5–2.5) with UPDRS (Unified Parkinsons Disease Rating Scale) Part IIImotor scores at a mean of 22.3 ± 7.8 (Table 2.1). All PD subjects weretested in the off-medicated state after a 12-hour overnight withdrawal fromL-dopa medication. Other medications that some subjects were on included:amantadine, ramapril and atorvastatin.Table 2.1: Demographical data of the PD subjectsPatientnumberAge(Year)Sex Duration(year)UPDRSIIIHoehn &YahrHandedness1 58 M 4 18 2 R2 64 F 4 12 1.5 R3 67 M 4 16 2 R4 56 M 2.5 21 2 L5 53 M 3 32 2.5 R6 49 M 7.5 35 2 R7 65 F 5 32 2 R8 68 M 1.5 22 2 R9 66 M 1 24 2 R10 70 M 1 21 2 R11 59 M 1.5 10 2 R12 62 M 3.5 24 2 R282.2. Materials and MethodsFigure 2.1: Behaviour Task. (A) Subjects faced a screen with a target (blue) that movedvertically up and down, and controlled a cursor (yellow) using a joystick. The errordifference (∆) between the actual positions of the target and cursor was amplified by ascaling factor (α): ∆ × α = displayed visual error feedback (B) Trials (90 s) alternatedbetween ‘Better (B)’ and ‘Worse (W)’ conditions.2.2.2 Ethics StatementThe study was approved by the University of British Columbia ClinicalResearch Ethics Board. All subjects gave written, informed consent prior toparticipation. Research was conducted according to the principles expressedin the Declaration of Helsinki.2.2.3 Visuomotor Tracking TaskSubjects were comfortably seated 80 cm in front of a screen and performeda manual tracking task. On the screen, a target (blue) and cursor (yellow)connected by a black horizontal rod were displayed (Fig. 2.1). The targetbox oscillated vertically up and down with the summation of two frequencies(0.06 and 0.1 Hz). Subjects controlled the cursor using a joystick with theobjective of matching the horizontal position of the cursor to the target– i.e., to keep the horizontal black rod straight. The tracking error (∆,difference between the actual positions of the target and cursor) was scaledby a factor (α) to determine the displayed position of the cursor: ∆ × α= displayed visual error feedback. In the ‘Better (B)’ task condition, αwas set to 0.3, and in the ‘Worse (W)’ task condition, α was set to 2, suchthat it artificially appeared to subjects that they performed better or worse,respectively, based on their scaled error feedback.During the experiment, subjects performed a total of 8 trials. Each trial(90 s) was comprised of three alternating blocks (30 s each) of B and Wconditions – with Trial 1 ordered as B-W-B and Trial 2 ordered as W-B-W(Fig. 2.1). During each trial, either a subthreshold verum current (90%292.2. Materials and Methodsof cutaneous sensory threshold) or sham current stimulation was delivered.Four trials contained verum EVS delivery whereas the other four trials con-tained sham stimulation. Subjects were unaware of either verum or shamstimulation since the order in which stimuli were delivered was pseudoran-dom, and the verum stimulation was imperceptible to the subject. Eachtrial was followed by a break (30 s) to preclude a hysteretic effect carryingover to the next trial. Before starting the experiment, subjects were allowedto practice tracking the target and using the joystick as needed in at leastone practice trial. Due to technical details of the data capture system, thecursor position was irregularly sampled at ∼55 Hz. We then resampled thedata at exactly 50 Hz using linear interpolation before further analyses.2.2.4 StimulusEVS was delivered to subjects through carbon rubber electrodes (17 cm2)in a bilateral, bipolar fashion. For bilateral stimulation, an electrode wasplaced over the mastoid process behind each ear, and coated with Tac gel(Pharmaceutical Innovations, NJ, USA) to optimize conductivity and adhe-siveness. The average impedance of the subjects was measured around 1 kΩ.Digital signals were generated on a computer using MATLAB and convertedto analog signals via a NI USB-6221 BNC digital acquisition module (Na-tional Instruments, TX, USA). The analog command voltage signals weresubsequently passed to a constant current stimulator (Model DS5, Digitimer,Hertfordshire, UK), which was connected to the stimulating electrodes.Bipolar stimulation signals were zero-mean, linearly detrended, noisycurrents with a 1/f -type power spectrum (pink noise) as previously appliedto PD and healthy subjects [278, 357, 432]. The stimulation signal wasgenerated between 0.1–10 Hz with a Gaussian probability density, with thecommand signal delivered to the constant-current amplifier at 60 Hz (Fig.2.2). The stimulus was applied at an imperceptible level to avoid effectsby general arousal and/or voluntary selective attention, with the currentlevel individually determined according to each subjects cutaneous sensorythreshold.Since perception of EVS is inherently subjective, we utilized systematicprocedures that have been previously used in determining subliminal cur-rent levels for both EVS and transcranial stimuli [154, 393, 422]. Startingfrom a basal current level of 0.02 mA, noisy test stimuli were delivered for20 s periods with gradual stepwise increases (0.02 mA) in current intensityuntil subjects perceived a mild, local tingling in the area of the stimulat-ing electrodes. As performed previously, a threshold value was defined once302.2. Materials and MethodsFigure 2.2: Characteristics of the stimulus. (A) Typical recording from a subject receivinga noisy stimulus applied for 90 s duration. The stimulus presented is at the highest currentintensity (current level 6), which is set to 90% of the subjects individual sensory threshold(RMS current value of 0.266 mA). (B) Probability density function of the stimulus currentfollows a Gaussian distribution.subjects reported a tingling sensation [393, 422], which lasted for the dura-tion of the test stimulus. The current level was then decreased each timeby one level until sensation was no longer reported during delivery of teststimulus pulses, and increased by one step in current intensity to confirmthreshold. Each delivery of a test stimulus was followed by a period of nostimulation for at least 30 s to preclude a hysteretic effect carrying overto the next test stimulus. Subjects were blind to the onset and durationof test stimuli, as well as the threshold-testing scheme. After completingthe threshold test and throughout the experiment, stimuli were deliveredat subthreshold intensity (0.19–0.90 mA), which is achieved at 90% of thedetermined cutaneous sensory threshold value.2.2.5 Behavioural Data AnalysisWe employed both exploratory and hypothesis-driven analysis methods toanalyze the behavioral data. We initially analyzed the data on a subject-by-subject basis as we were unclear whether or not there would be substantialintersubject variability to EVS response. LDA was first used to see if track-ing behavior could be reliably discriminated depending upon whether EVSwas applied or not. We derived a EVS linear discrimination function, g(X),to create maximum separation between means of the projected classes with312.2. Materials and Methodsminimum variance within each projected class:g(X) = w1X1 + w2X2 + ...+ w21X21 + ω0 = wtXt + ω0 (2.1)where X = [X1 X2 ... X21] ∈ Rn×d is a data matrix of n d-dimensional sam-ples in which each column represents an independent variable, w= [w1, w2,..., w21] ∈ Rd×1 the weight vector containing linear coefficients of the vari-ables in the data matrix X, and ω0 the bias-weight. LDA was applied tothe “Better” and “Worse” conditions separately.For this exploratory part of the analysis, we included linear (first-order)and non-linear (second- and third-order) combinations of variables in theEVS discriminant function (Table 2.2). During the experiment, we variedthe phase of the initial target trajectory not only between subjects but alsobetween the trials to prevent the subjects from easily predicting upcomingtarget movement. Therefore, variables from X1 to X9 were included asnuisance variables in the LDA to account for the target differences.Table 2.2: Variables in LDA modelNotation Variables1X1, X2, X3 T (t), T (t)2, T (t)3X4, X5, X6 VT (t), VT (t)2, VT (t)3X7, X8, X9 AT (t), AT (t)2, AT (t)3X10, X11, X12 D(t)− T (t), {D(t)− T (t)}2, {D(t)− T (t)}3X13, X14, X15 VD(t)− VT (t), {VD(t)− VT (t)}2, {VD(t)− VT (t)}3X16, X17, X18 D(t+ ∆t)−D(t), {D(t+ ∆t)−D(t)}2, {D(t+ ∆t)−D(t)}3X19, X20, X21 VD(t+ ∆t)−VD(t), {VD(t+ ∆t)−VD(t)}2, {VD(t+ ∆t)−VD(t)}31 T=target position, VT=target velocity, AT=target acceleration, D=displayed cursorposition, VD=displayed cursor velocity, t=time index, and ∆t=reaction delay of 0.5 s[168]To test for significance of the LDA results, we employed bootstrappingtechniques. We permuted the EVS labels (on/off) and then re-computed theLDA function with the permuted data. This was repeated 1000 times. Anyweight value from the original LDA function g(X) whose absolute value322.2. Materials and Methodswas greater than all the weights computed from the permuted data wasconsidered to be significantly influenced by EVS.In addition, a multivariate linear regression model was used to test thehypothesis that EVS had a significant effect on cursor position during track-ing. As the traditional least squares regression may be sensitive to noisy andgross errors [5], we chose a robust regression method to analyze our data(“robustfit” function in MATLAB). This method is known to be robust tooutliers utilizing an iteratively reweighted scheme to deweight the influencesof outliers. With cursor position as a response variable (Yi), the followingregression model was proposed:Yi = Aiβ + i (2.2)where for each data point i we have the vector of independent variablesAi = [Ai1, ..., Ai5], the vector of regression coefficients β solved by a bisquareweighting function, and the residual i (assumed to be independent andidentically distributed Gaussian). The selected independent variables aresummarized in Table 2.3 (note that A1, A2 and A3 are same as the variablesX1, X4 and X10 in Eq. 2.1, respectively). The categorical variable of EVSwas denoted with either 0 (EVSoff) or 1 (EVSon). We tested for significanceof the coefficients under the null hypothesis that the coefficient estimateswere equal to zero.For a signal-to-noise ratio (SNR) analysis, we utilized “snr” functionin MATLAB to calculate SNR of cursor trajectories. This examines thefundamental frequencies of the tracking trajectory plus the next 6 harmonics,and assumes that any power in the spectrum than these peaks are “noise”.Table 2.3: Estimated coefficients in the robust regression model (Eq. 2.2) and the P valueVariables (A) Coefficient estimates (β) P valuetarget position (A1) 1.00 0.0000target velociy (A2) -0.0779 0.0000displayed cursor position − target position (A3) 0.501 0.0000cursor velocity − target velocity (A4) -0.0160 0.0002EVS (A5) 3.99e-05 0.0410R2=0.8811332.3. ResultsFigure 2.3: Coefficients of the variables of the linear discriminant function in the Worsecondition. The x-axis represents variables from X10 to X21 in Table 2.2 while the y-axisrepresents weight (w) value. The computed coefficients are depicted as black for the EVSdiscriminant function and blue for bootstrapping. Red asterisks denote coefficients thatare outside the 95% confidence interval of bootstrapping.2.3 Results2.3.1 Results of LDA in Worse ConditionCoefficients of EVS discriminant function (Eq. 2.1) were calculated for eachsubject and are plotted as black lines in Fig. 2.3. For clarity, nuisancevariables related to absolute target position (i.e., X1 −X9) are not shown.The 1000 sets of linear coefficients generated from the bootstrapping aredepicted as blue lines. In most subjects, the coefficients w10, w11 and w12 ofg(X) (representing linear and higher powers of the perceived error betweenthe target and the displayed cursor position) were robustly modulated byEVS. In addition, displayed cursor velocity (w16 or w17) and acceleration(w19, w20 or w21) were also found to be significantly affected by EVS acrosssubjects.2.3.2 Results of LDA in Better ConditionFig. 2.4 shows the LDA results in the Better condition. As before, coeffi-cients w10, w11 and w12 were significant among all the subjects. In addition,10 out of 12 subjects showed significant w18 weightings. Other coefficientswere not robustly seen in all subjects. For example, unlike the LDA results342.3. ResultsFigure 2.4: Coefficients of the variables of the linear discriminant function in the Bettercondition. The x-axis represents variables from X10 to X21 in Table 2.2 while the y-axisrepresents weight (w) value. The computed coefficients are depicted as black for the EVSdiscriminant function and blue for bootstrapping. Red asterisks denote coefficients thatare outside the 95% confidence interval of bootstrapping.of the Worse condition, displayed cursor acceleration (w19, w20 or w21) wasno longer significantly influenced by EVS in the Better condition.2.3.3 Results of Robust Regression ModelTable 2.3 is the coefficient estimates of the variables of the multivariateregression model (Eq. 2.2) and their P values. The computed R2 of theregression model was 0.8811. EVS was significantly associated with cursorposition across all subjects (P < 0.05).2.3.4 Effect of EVS on Cursor OvershootingIn order to get an intuitive interpretation of EVS effects, we calculated theEVS discriminant function values (Eq. 2.1) for each subject. We used datafrom trials 1 and 7 for the calculation as these two trials had identical phasesof the trajectories, with a difference in whether or not EVS was delivered(EVSon for trial 1). Then, ∆g was computed by subtracting the functionvalues of trial 7 from trial 1. By plotting ∆g, we could not only locate EVSeffects on the cursor trajectory but also directly make visual comparisonof the cursor movement in the identified location. Fig. 2.5 shows target352.3. Resultstrajectory, cursor trajectory and ∆g for each subject.Figure 2.5: Trajectories of target (blue) and cursor (EVSon: red, EVSoff: black) and∆g (black bar in the bottom). ∆g was computed by subtracting the linear discriminantfunction values of trial 7 (EVSoff) from trial 1 (EVSon).The trials alternated betweenW-B-W conditions (each condition 30 s).The effect of EVS was greatest near sinusoidal peaks. This trend wasfound in most of the subjects regardless of how well the subjects tracked thetarget. For instance, subject 5 tracked the target relatively better comparedto the other subjects, and ∆g was significant around at 5, 20, 65, and 80 s.Subjects 11 and 12 performed the tracking task poorly, but the EVS effectsstill appeared near sinusoidal peaks.One of the noticeable features on the peaks is a degree of overshooting ofcursor trajectories. To assess a possible relationship to EVS stimulation,we compared the difference between the cursor position and the target onthe peaks. Fig. 2.6 shows a representative example of cursor overshootingnear sinusoidal peaks in target. The peaks in cursor appeared with somelagged time (∆t). The amplitude of the target peaks was subtracted fromthe cursor peaks, and the difference (∆d) was defined as cursor overshooting.Cursor peak was defined when the cursor position was at its max/min point.Cursor overshooting was calculated for all trials and subjects, then averageddepending on the task conditions and presence of EVS stimulation as shownin Table 2.4. The P value was calculated from ANOVA of the means betweenEVSon and EVSoff (i.e., a single, two-level factor).In Worse condition, the subjects tended to overshoot significantly less362.3. Resultson the lower peaks while stimulated by EVS. On the upper peaks, the meanovershooting of EVSon was also smaller than EVSoff, but the differencewas not significant. In Better condition, however, there was an increasingtendency for cursor overshooting with stimulation.Figure 2.6: Representative example of cursor overshooting on upper and lower peaks fromSubject 1. Cursor overshooting (∆d) was calculated as cursor position – target position.∆t represents time difference between peaks in cursor and target trajectories.Table 2.4: Means of cursor overshooting on sinusoidal peaks and ANOVA resultsLower peak Upper peakEVSon EVSoff P value EVSon EVSoff P valueWorse -0.0517 -0.0714 0.0036 0.0695 0.0784 0.22Better -0.0946 -0.0451 0.0038 0.0890 0.0690 0.142.3.5 Effect of EVS on SNR of Cursor TrajectoryMovement variability is another important feature to characterize the track-ing performance. Particularly, in goal-directed behavior, the variability orig-inates from collateral movement to the main goal of a task. In this sense,the cursor trajectories in our tracking test can be seen to a combination oftwo components. One is the primary movement whose form is similar to the372.4. Discussiontarget trajectory, and the other is submovement that may appear as noisesuperimposed on the primary movement. In order to investigate if EVS hadaffected movement variability of the subjects, we calculated SNR of cursortrajectories and compared differences in between EVSon and EVSoff con-ditions. As shown in Fig. 2.7, the mean SNR of 12 PD subjects was 27.6when EVS was applied, which was significantly greater than 21.3 in EVSoffcondition (P < 0.05).Figure 2.7: Comparison of SNR of cursor trajectories between EVSon and EVSoff condi-tions.2.4 DiscussionOur results demonstrate that noisy EVS robustly influences motor trackingperformance in PD patients off dopaminergic medication. Motor improve-ments are consistent with results previously reported in hemiparkinsonianrats [331] whereby EVS with a 1/f power density improved rod performance.Previously, we demonstrated that noisy EVS has the ability to modulate syn-chronization of broadband EEG oscillations in healthy subjects [177]. Ourrecordings of EEG rhythms were observed at resting-state, suggesting thatnoisy EVS was able to modulate cortical activity and presumably connectedsubcortical-cortical projections. In this study, we observed a functional effectof EVS on sensorimotor processing and motor performance in a visuomotortask, suggesting that noisy vestibular stimulation modulates motor networksin PD subjects.Our results seem to indicate that noisy EVS affects the sensitivity ofmotor responses (in this case, joystick-controlled cursor position) to visual-ized error (displayed cursor position – target position). We do not believe382.4. Discussionthat our observed results are the consequence of an attentional or generalarousal effect, such as through activation of the reticular activating system.The imperceptible nature of our stimulus, which subjects were not aware ofthroughout the experiment trials, precludes this issue which is present withother forms of minimally invasive stimulation methods [118].Depending on the stimulus parameters (i.e., current intensity, frequency,signal shape), EVS is known to induce a broad range of effects, includ-ing eye movements, postural control and movements [111]. Therefore, oneinterpretation of our results may include the confounding effects of nystag-mus and/or ocular torsion through activation of the vestibulo-ocular reflex(VOR) [437]. Since subjects rely on visual error feedback, ocular torsionwould potentially hamper the perceived error feedback through a subjectivetilt in the visual perceptual field [437]. However, we note that our stimuluslevels were weak, whereas the preferred EVS current intensities for inducingocular torsion and subsequent perceptual tilts through EVS are much higher— at around 1–3 mA [437]. Therefore, we presume that our subthresholdstimulus was not strong enough to notably induce confounding visual effectsand corollary perceptual changes in our experiment.Noisy EVS is known to modulate EEG spectral power. Wilkinson et al.have demonstrated that noisy EVS is able to modulate the EEG spectralpower during a face processing task [422]. Our previous study has demon-strated that noisy EVS is able to modulate the EEG synchrony patterns inhealthy subjects [177]. Altogether, these findings combined with our presentresults suggest that noisy EVS is able to modulate oscillatory activity inresting and task-related networks, which involve sensorimotor processing inour particular study. The motoric effects of EVS may be related to modula-tion of oscillations related to integration of information and error-processing.Since perceived error (i.e., the error between the target and the displayedcursor position) was robustly detected by the LDA analysis, fronto-midline(FM) theta may be a candidate oscillation to be modulated by EVS in PDsubjects. FM-theta shows an increased amplitude during tasks requiringconcentration [252], which is related to error-related negativity (ERN), anevent-related potential seen after errors are made. FM-theta may representa universal mechanism for action monitoring with the midcingulate cortexacting as hub for the integration of information [63]. Thus, our results sug-gest that EVS may regulate FM-theta activity in PD subjects.The increased SNR shown in Fig. 2.7 suggests that application of noisyEVS may have increased synchronization in neuromotor system via stochas-tic facilitation. Stochastic facilitation is a term to describe phenomena wherestochastic biological noise elicits functional benefits in a non-linear system392.4. Discussionsuch as the nervous system [238]. Several studies have reported that apresence of additive noise allows a weak input signal to be better detected,resulting in an increase in SNR in EEG [119, 180, 231, 361, 375, 402, 413]and sensorimotor performance [246]. These findings suggest that noisy EVSinput may also be able to modulate detection and transmission of the sen-sorimotor system via stochastic facilitation, resulting in an increase in syn-chronization of the neuromotor system. However, a further investigation isrequired to elucidate whether the synchronization is limited to cortical areasor if it could give rise to corticomuscular synchronization [246].We further speculate that our results may be at least partly explainedby modulation of cortico-BG rhythms involved in sensorimotor processing.Growing observations suggest a concept that the BG regulates action moti-vation or response ‘vigour’ [265, 327] as well as the speed and size of move-ment [360, 380]. Deficient scaling of the initial burst of earliest agonistmuscle activity (EMG) to meet the demands of a motor task is frequentlyobserved in clinical disorders of the BG, such as PD. The link between mo-tivation and movement gain may be universally weakened in Parkinsoniansubjects [20, 380]. We thus speculate that EVS may also correct deficientvigour caused by BG dysfunction through modulation of pathological brainrhythms.We note that we used a single noisy stimulus for all subjects. However,the results shown in Fig. 2.3 also emphasize the importance of looking atpatient-specific stimuli. For instance, the coefficients regarding the differencebetween cursor and target velocities (w13, w14, and w15) were found to besignificant in some subjects, but were indistinguishable from bootstrappingfor the rest of the subjects.Finally, we note that EVS had fewer effects in the Better condition com-pared to the Worse condition. Presumably, subjects would have made fewercorrective movements in the former condition. This raises the possibilitythat EVS may also depend upon the number and form of corrective sub-movements. As submovements were not captured by the global LDA andmultivariate regression methods used here, this warrants further investiga-tion.40Chapter 3Quadrature regression andIVA approach to removal ofhigh-voltage EVS artifactsfrom simultaneous EEGrecordingsChapter 2 demonstrated the robust effects of noisy EVS on the motor track-ing performance in PD. The positive effects on movement suggest intriguingquestions how EVS modulates upstream neural activities in the brain thatcontrol downstream movement. One way to investigate this is to record cor-tical activities via EEG while delivering EVS. However, simultaneous EEGand EVS studies have been hindered by the high-voltage stimulation arti-facts that completely distort the EEG signals. In this chapter, we tackledthis problem by introducing a novel denoising method.3.1 IntroductionOver the last decade, noninvasive electrical brain stimulation (NEBS) hasbeen extensively explored as a means of studying fundamental mechanismsunderlying cognitive and motor functions, as well as a potential therapy forneurological diseases. NEBS techniques such as tACS and EVS can modu-late ongoing neural oscillations, which play a fundamental role in brain func-tioning [390]. While NEBS studies are often conducted with brain imagingmodalities such as fMRI and positron emission tomography (PET), it is theEEG that is particularly valuable for NEBS studies as it is relatively inex-pensive, potentially portable, and able to record the electrical brain activitynoninvasively with high temporal resolution.Identifying ways to properly analyze the EEG acquired during stimula-tion has been a major challenge to date in fully understanding mechanisms413.1. Introductionof NEBS. In general, the electrical artifacts induced by NEBS are so dis-ruptive that groups have resorted to avoiding the artifact removal problementirely by simply comparing EEG in pre- and post-stimulation conditions(for review, see [18] and [392]).Several factors contribute to the technical difficulties of removing NEBSartifact from EEG. First, the stimulation artifact amplitude is several ordersof magnitude larger compared to actual brain signals, often obscuring theEEG completely. Second, the frequency ranges of stimulation and corticaloscillations of interest can potentially overlap, precluding the application ofsimple signal processing techniques such as digital filtering. Third, stimu-lation artifacts recorded downstream in the EEG do not exactly match thedelivered currents at the stimulation electrodes, differing in phase-shifts,amplitude variations and other morphologic alterations in the waveforms.It is likely that the applied stimulation currents go through a non-lineartransformation before being recorded at EEG electrodes due to resistive andcapacitive effects present at the interfaces between EEG electrodes, gel, andskin layers and possibly a non-stationary transformation due to electrodeimpedances changing over time. Fourth, the ground truth of ongoing brainresponses to stimulation is unknown, making it difficult to determine thesuccess of methods proposed to disentangle stimulation artifacts from ongo-ing brain activity. Therefore, to be able to adequately tackle the stimulationartifact rejection problems, a thorough investigation of the characteristics ofstimulation artifacts, exploration of more advanced analytical methods, andadequate assessment of the performance of different methods are required.A few methods have been proposed to remove large-amplitude artifactsin EEG signals. Most of the methods mainly stem from EEG-fMRI studieswhere EEG signals are severely corrupted by artifacts caused by switch-ing of the magnetic field gradient during MR acquisition [8]. Similar toNEBS, the artifact can have an amplitude a few hundred times greater thanthe EEG signals. A moving average algorithm is the most commonly-usedmethod to remove this artifact, whereby an artifact template is createdfrom an average of several adjacent time windows (or trials) and then sub-sequently subtracting this mean template from the raw EEG signal. A sim-ilar variant of this approach has been applied in recent EEG-NEBS studies[25, 147, 187, 321, 407] (Table 3.1). Since this approach frequently leavessmall remnants of artifact, an additional step is usually taken to separateany remaining artifacts from brain activity using methods such as princi-pal component analysis (PCA) or independent component analysis (ICA).However, the moving average algorithm can fail in removing artifacts whensignificant phase-shifts and/or amplitude changes are present between time423.1. IntroductionTable 3.1: Artifact Rejection methods used in EEG-NIBS StudiesStimulus Methods Data used PerformanceevaluationtDCS High-pass filtering (>2 Hz) fol-lowed by ICA• Real EEG dataduring transcranialdirect current stim-ulation (tDCS)• Manual inspection of IC5,10and40 HztACSArtifact template subtractionfor each channel• Real EEG dataduring tACS• Comparison of powerspectral density in the al-pha band during tACS at5, 10, and 40 Hz• Comparison of individualalpha frequency changesdue to eye closure duringstimulation to those in thesham condition10 HztACSTwo-step procedure1) Artifact template subtraction2) PCA for remaining artifacts• Simulated EEG• Real EEG dataduring tACS• Comparison of meanspectral power before/afterartifact rejection• Physiological phase-dependent response to thevisual stimulus2, 6, 12,25, 40,70 and100 HztACSTwo step procedure1) Subtraction from each EEGchannel a properly scaled andphase-shifted fraction of the sumof the TP9 and TP10 that arenear the tACS electrodes2) Digital notch filter at therespective stimulation frequencyand the first two harmonics• Real EEG dataduring tACS• Demonstration of on-line effects in lower gammabandwindows. Additionally, if a fraction of neural responses is phase-locked tothe stimulation, they will tend to be removed when the average template issubtracted from the EEG signals.A recently-proposed joint blind source separation (JBSS) technique hasbeen successfully applied to EEG signal denoising applications. For ex-ample, gradient artifacts from EEG-fMRI can be more robustly removedby independent vector analysis (IVA) compared to the artifact subtractionmethod described above [2]. IVA has also shown to be effective in removingmuscle artifacts in real ictal EEG data and outperforms ICA in isolatingboth ocular and muscle artifacts [64, 66].In this chapter, we propose utilizing JBSS approaches to separate NEBS433.2. Materials and Methodsartifacts from brain signals by using multiset canonical correlation analysis(MCCA) and IVA. To the best of our knowledge, this is the first attempt toinvestigate JBSS approaches in NEBS artifact removal. In contrast to PCAand ICA, which decomposes a single dataset into individual components,these relatively new methods simultaneously accommodate multiple datasetsand can extract underlying common sources from the signals (for a technicalreview of the methods, see [67]). By jointly analyzing multiset data, MCCAand IVA identify source components that are maximally correlated acrossdatasets yet constrained to be uncorrelated (or in the case of IVA, maximallyindependent) within a dataset. Artifact-corrupted EEG can be segmentedand restructured into multiple sets (i.e., epochs) based on the period ofstimulation signal or repeated trials. As the stimulation artifacts in EEGpossess relatively similar (but not identical) amplitudes and phases fromepoch-to-epoch and are minimally correlated or statistically independentfrom brain activities, our hypothesis is that JBSS approaches would result insuperior performance in isolating stimulation artifacts as source componentscompared to conventional artifact rejection methods. Reconstruction of theEEG without artifact component(s) would then results in a “cleaned” EEG.This chapter is organized as follows: Section 3.2 describes the exper-iment setup and protocol, EEG data acquisition and preprocessing, andtechnical details of five different methods tested to remove stimulation ar-tifacts: PCA, second-order blind identification (SOBI), MCCA, IVA, andquadrature regression-IVA (q-IVA) proposed here. The methods were testedthrough both simulation and real EEG data, and Section 3.3 compares theperformance of the artifact removal methods, followed by discussions andsuggestions for future work in Section 3.4.3.2 Materials and Methods3.2.1 SubjectsThirteen healthy people (6 females, age=64.9 ± 15.7) participated in thestudy (Table 3.2). No subjects had any reported vestibular or auditorydisorders.The experimental protocol was approved by the Clinical Research EthicsBoard at the University of British Columbia. All subjects gave a writteninformed consent before the beginning of the experiment.443.2. Materials and MethodsTable 3.2: Subjects informationSubject ID Data usage1 Simulation study (simulated stimulation artifacts)2 Simulation study (resting EEG)3-12 Real data study (performance validation using Pdiff )13 Real data study (performance validation using changes in alpha power by eye closure)3.2.2 EVSEVS was delivered in bilateral, bipolar fashion through pre-gelled Ag/AgClelectrodes (BIOPAC Systems Inc., CA, USA) placed over the mastoid pro-cess behind each ear. The EVS signal was generated on a computer usingMATLAB (MathWorks, MA, USA) software and converted to analog signalsthrough a NI USB-6221 BNC digital acquisition module (National Instru-ments, TX, USA). The analog command voltage signals were subsequentlypassed to a constant current stimulator DS5 (Digitimer, UK). A multisinesignal in the theta (4–8 Hz) frequency band was used as the EVS stimulus(Fig. 3.1). The frequencies of sinusoids were uniformly distributed every 0.2Hz and the phases were chosen to minimize the crest factor by a clippingalgorithm [401]:x(t, φ) = a ·n∑i=1cos(2pifit+ φi) (3.1)where x(t, φ) is the multisine stimulus, a is the amplitude of the mul-tisine, fi and φi are the frequency and phase, and i is the index of eachsinusoidal component (f1, f2, f3, ..., fn = 4.0, 4.2, ..., 8.0 Hz).The stimulus was applied at an imperceptible level to avoid effects by generalarousal and/or voluntary selective attention, with the current level individ-ually determined at 90% of each subjects sensory threshold (see 3.2.3 StudyProtocol).3.2.3 Study ProtocolSince individuals have inherently subjective perception of EVS, we utilizedsystematic procedures that have been previously used in determining sub-liminal current levels [177, 205]. For each subject, the multisine stimulus453.2. Materials and Methodswas delivered at a basal current level of 0.05 mA for a period of 10 secondsand the level was increased stepwise (0.02 mA) until the subject perceiveda mild, local tingling sensation in the area of the electrodes. The currentlevel was then decreased by one level each time until sensation was no longerreported during delivery of the stimulus, and increased by one level to con-firm threshold. The measured individual threshold level was in the range of0.31-0.77 mA.After the threshold had been determined, the subjects were comfortablyseated 80 cm from a screen and were instructed to focus their gaze on acontinuously-displayed fixed target to minimize distractions while EEG wasrecorded. EEG was first recorded without stimulation for 10 s (pre-EVS),blinding subjects to the actual stimulus onset. The multisine stimulus wasthen delivered for 60 s consisting of 6 consecutive trials. In each trial, EVSwas on for 5 s (during-EVS) and off for 5 s (post-EVS) (Fig. 1). For SubjectFigure 3.1: Experimental setup and an overall flow diagram for the study. (a) Placement ofEEG (yellow) and EVS (red and black) electrodes, and 5-s EVS stimulus. (b) The stimuluswas delivered for 60 s with 6 trials of 10-s epochs consisting of 5-s EVSon and 5-s EVSoff.To illustrate relative scales of the stimulation artifacts to EEG, sample traces (channelsC3, C4, O1 and O2) from one subject are shown. The EEG data were preprocessed andEVSon segments were formed into M sets of K × T matrices in order to apply JBSSmethods (i.e., MCCA, IVA and q-IVA).13, we also measured resting EEG with the eyes open (60 s) and closed (60s) in the beginning of the study. Then, the subject performed the studyprotocol with the eyes open followed by a 1-min break, repeating it with theeyes closed.463.2. Materials and Methods3.2.4 EEG Data Acquisition and PreprocessingEEG was recorded from 27 scalp electrodes using a Neuroscan SynAmps2EEG acquisition system (Neuroscan,VA, USA) and a standard electrode cap(64-channels Quik-Cap, Neuroscan, VA, USA). EEG electrodes were posi-tioned according to the international 10-20 placement standard with oneground and one reference electrode, and two earlobe electrodes were placedon each side for re-referencing purposes. The electrodes were attached usingElectro-Gel (Electrode-Cap International, OH, USA) and impedances werekept below 10 kΩ. All signals were sampled at 1 kHz, and no clipping wasobserved during stimulation. The EEG data were bandpass filtered between3 and 55 Hz using a two-way finite impulse response (FIR) filter (the eegfiltfunction in EEGLAB [84]), and re-referenced to the average reference (linkedearlobe) offline. Ocular artifacts (EOG) in the EEG recorded while stimu-lation was off were corrected based on cross-correlation with the referenceEOG channels using the AAR toolbox included in EEGLAB.3.2.5 Simulation DataSimulations were performed in order to quantitatively assess and compareperformance of different artifact rejection methods. Simulation data werecreated by combining the resting (i.e., artifact-free) EEG data from Subject2 with simulated EVS artifacts that were obtained by fitting an electri-cal circuit model to the EEG recorded from Subject 1. We note that inprevious studies [147], stimulation artifacts were created simply by addingtime-jitter to the stimulation model with amplitude weighted differently foreach channel depending on the distance from stimulation electrodes (i.e.,the artifact amplitude was largest for adjacent channels and gradually de-creased with distance further away from the tACS electrodes). While thisaccounts for some of the characteristics of stimulation artifact such as timelags caused by mismatch of internal clocks of EEG recording and stimulationsystem, it does not capture the characteristics caused by changes in bodyimpedance. For example, even respiration and heart beats result in headand body movement that slightly change the distance between stimulationcurrent and EEG sensors, modulating the electrode-tissue impedance [268].Therefore, instead of the conventional method, a resistive-capacitive circuitmodel for the physical electrode-skin interface [3] was adopted to generatesimulated EVS artifacts (Fig. 3.2):473.2. Materials and MethodsFigure 3.2: The electrical circuit model for the physical electrode-skin interface adaptedfrom [3] (Ehe: the half cell potential of the electrode/gel interface; Rd and Cd: the resistiveand capacitive components of the impedance associated with the electrode/gel interface;Rs: the series impedance associated with the resistance of the electrode gel; Ese: thepotential difference across the epidermis; Re and Ce: the resistance and capacitance ofthe epidermis; Ru: the resistance of the dermis and subcutaneous layer.Z(s) ≈ ( RdsRdCd + 1+ResReCe + 1‖ RpsRpCp + 1) ≈ b1s+ b0s2 + a1s+ a0(3.2)where Rd and Cd are the the resistance and capacitance associated with theelectrode-gel interface, Re, Ce, Rp and Cp are the resistance and capacitanceassociated with a skin structure consisting of epidermis, dermis, and a sub-cutaneous layer, s is the complex frequency variable, and a0, a1, b0 and b1are coefficients in the transfer function [3].We thus modelled the electrode-skin impedance structure as a second-order, continuous-time transfer function with one zero and two poles. Theprocess of generating simulated EEG data using (Eq. 3.2) is described inFig. 3. For illustrative purposes, an example of one channel is shown insteadof 27 channels. Using the system identification toolbox in MATLAB withthe multisine signal as the input and the EEG signal recorded from Subject1 as the output, the coefficients in (Eq. 3.2) were obtained (Fig. 3.3(a)).To ensure robustness of results, a small amount of random variation wasadded to the obtained coefficients to generate 600 simulated artifacts (=6 epochs × 100 realizations) so that each artifact had a small phase and483.2. Materials and Methodsamplitude variation from the input signal (Fig. 3.3(b)). Specifically, foreach realization in (Eq. 3.2), we modelled a0 ∼ N (0.6, 0.1), a1 ∼ N (10.8, 1),b0 ∼ N (249.6, 10), and b1 ∼ N (7155.5, 100), where N (µ, σ) refers to aGaussian distribution with mean and standard deviation. The amplitudewas then scaled to obtain Ys,EV S so that its maximum peak matched withthe one of the real artifact in Fig. 3.3(a). Ys,EV S was then superimposedon the resting EEG data from Subject 2, Ys,EEG, to create the simulationdata, Ys:Ys = Ys,EEG + Ys,EV S (3.3)Figure 3.3: Example of generating simulated EEG data, Ys. (a) The parameters (a0,a1, b0 and b1) of the second order transfer function in (Eq. 3.2) were estimated usingthe multisine signal, u(t), as the input, and stimulation artifacts recorded in EEG as theoutput. (b) New parameters (a′0, a′1, b′0 and b′1) were obtained by adding a small amountof random variation to the original parameters. The new parameters were then used togenerate simulated artifact, y′(t). (c) Ys,EV S was created by scaling y′(t) to match themaximum peak value in the raw EEG data. The final simulated EEG data, Ys, was createdby adding Ys,EV S to the resting EEG from another subject, Ys,EEG.493.2. Materials and Methods3.2.6 Artifact Rejection MethodsPCA and ICAICA is a method to find statistically independent sources from mixed signalsby using higher-order statistics. Given minimal prior information about theunderlying sources as well as the mixing process, ICA decomposes observedsignals and finds underlying sources such that every source is independentof the others. ICA has been widely utilized in EEG studies to identifymeaningful neurophysiological signals and separate them from a wide varietyof artifactual sources [86]. The basic premise for using PCA is that the largeamplitude of stimulation artifacts would account for such a large variationof the recorded EEG that the artifacts would not co-vary with (i.e., beorthogonal to) brain activity. For ICA, the assumption is that EEG datarecorded from scalp electrodes are considered linear summations of brainactivity and stimulation artifacts that are statistically independent fromeach other. We used a commonly-used ICA algorithm, SOBI [26, 200].To account for possible non-stationary relations between electrodes acrossepochs, PCA and SOBI were applied to each epoch separately rather thanto a single matrix created by concatenating all 5-s EEG epochs.MCCA and IVACCA identifies canonical variates from two multidimensional variates bymaximizing correlation between them. Given two random vectors x1 andx2, CCA finds two transformation vectors, a and b, such that the canonicalvariates, y1 = aTx1 and y2 = aTx2 have maximum correlation. After thefirst pair of canonical variates is found, the second pair of transformationvectors is obtained by deflation, so that the next corresponding canonicalvariates have maximum correlation with each other while still being uncor-related with the first pair of canonical variates [434].MCCA is an extension of CCA that allows for the joint analysis of morethan two data sets. The goal is to optimize an objective function to achievethe maximum overall correlation across the canonical variates. Since multi-ple correlations need to be considered, the MCCA algorithm takes multiplestages where one group of canonical variates is obtained in each stage byoptimizing the objective function to maximize the overall correlation [72].In the second stage of the algorithm, a constraint is applied such that theestimated canonical variates are uncorrelated with the previously obtainedcanonical variates. Among several cost functions proposed in [172], we usedthe MCCA procedure based on maximizing the sum of squared correlations503.2. Materials and Methods(SSQCOR) across the canonical variates.Similar to MCCA being an extension of CCA with capability of jointlyanalyzing more than two data sets, IVA is a generalization of ICA from oneto multiple data sets [67]. Applying ICA individually to each data set suffersfrom the permutation problem whereby the recovered source componentsfrom each data set may be inconsistently ordered across sets, resulting inambiguities of which source component in one data set is associated withother components across data sets. IVA addresses the permutation problemby assuming each source component within a dataset is related to a sourcecomponent in each of the other datasets as well as independent of all theother source components within the dataset [11, 179]. Several algorithmshave been proposed for IVA [352] so as to minimize the mutual informationbetween source components. Here, non-orthogonal IVA was used as it allowssource components following either multivariate Gaussian or non-Gaussiandistributions and does not restrict the demixing matrices to be orthogonalas in MCCA [11].q-IVAThe q-IVA method proposed here consists of two steps. Firstly, for eachepoch and channel separately, we remove high-amplitude stimulation artifactusing a regression model that includes the stimulation signal, x(t), and itsquadrature component, xˆ(t), to compensate for possible phase-shifts. For anarrowband signal like the theta-multisine stimulus used in this study, itsanalytical signal, z(t), can be expressed asz(t) = x(t) + jxˆ(t) (3.4)xˆ(t) = H[x(t)] =1pit· x(t) (3.5)where t is sampling time, x(t) is the multisine simulus, xˆ(t) is the quadraturecomponent (i.e., phase shift by −pi/2), and H is the Hilbert transform.For the EEG signal in channel k and epoch m, the regression model canbe written asyk(t)[m] = Xb[m]k + rk(t)[m] (3.6)X = [x(t), xˆ(t)] (3.7)513.2. Materials and Methodswhere k the channel index (k = 1, 2, ...,K;K = 27), m is the epoch index(m = 1, 2, ...,M ;M = 6), yk(t) is the T × 1 EEG signal, X is the T × 2matrix of the stimulation signal and its quadrature component, bk is the2 × 1 vector of regression coefficients, rk is the T × 1 vector of residuals,and T is the number of time points (T = 5000) in the epoch m. Taking theresidual of channel k in epoch m, rk(t)[m], in each row, six sets of residualmatrix were obtained:R[m] = [r1(t)[m], r2(t)[m], ..., rk(t)[m]]T (3.8)In the second step, we applied IVA to the residual matrices in order tofurther reduce the remaining artifact. IVA analyzes the residual matricesjointly and finds the estimated source components by minimizing the mu-tual information for all components and maximizing the mutual informationwithin each source component across the epochs [11]:R[m] = A[m] · S[m] (3.9)where A[m] is the K ×K mixing matrix, and S[m] is the K × T estimatedsource components matrix in the epoch . The cleaned EEG data were ob-tained by removing artifactual source components and projecting the restof the components back to the time domain. The number of the removedsource components was 2 (Fig. 3.4).3.2.7 Performance EvaluationSimulation StudyThe cleaned EEG data, Y˜s,EEG, were obtained by applying the aforemen-tioned five artifact rejection methods to the 100 realizations (N = 100)of the simulation data. With complete removal of the artifacts, Y˜s,EEG isexpected to be identical to Ys,EEG. Three measures were employed to eval-uate performance of the artifact rejection methods. As the first evaluationmeasure, relative root-mean-squared error (RRMSE) [65] of channel k wascomputed as the following:RRMSEk =1MM∑m=1RMS(y[m]s,EEG − y˜[m]s,EEG)RMS(y[m]s,EEG)(3.10)523.2. Materials and MethodsFigure 3.4: The principal components (PC) from PCA, underlying source components(IC) from SOBI, IVA and q-IVA, and canonical variates (CV) from MCCA.where y[m]s,EEG and y˜[m]s,EEG are T × 1 time series of channel k in Y [m]s,EEG andY˜[m]s,EEG matrices, and RRMSEk is the averaged value across all epochs forchannel k. The root mean square (RMS) for a time series vector y was533.2. Materials and Methodsdefined as,RMS(y) =√√√√ 1TT∑t=1y(t)2 (3.11)In order to measure the capability of preserving the original EEG signals,the correlation coefficient (CC) between and was calculated as the secondmeasure [65]. CCk was obtained by averaging the coefficient for channel kover all epochs:CCk =1MM∑m=1c[m]k (3.12)where c[m]k is the coefficient value in epoch m.As the third evaluation measure, power deviation (Pdev) was calcu-lated to investigate similarity of power spectral density between the cleanedand original EEG data. For each channel and epoch, the spectral powerwas calculated using Welch’s averaged, modified periodogram method (the“pwelch” function in MATLAB) and the average power in theta (4–8 Hz),alpha (8–13 Hz), beta (13–30 Hz) and gamma (30–55 Hz) bands was com-puted. For each frequency band, Pdev of channel k averaged over all epochswas computed as the following:Pdev,k =√√√√ 1MM∑m=1(P˜[m]k − P [m]k )2 (3.13)where P˜[m]k and P[m]k are the band power of Y˜s,EEG and Ys,EEG, respectively.Real Data StudyMCCA, IVA, and q-IVA were applied to the EEG data simultaneouslyrecorded during EVS from 10 subjects. Unlike the simulation approach, theground truth about “true” brain activities and remaining artifacts in thecleaned EEG data are unknown. We compared the power spectra betweenthe cleaned and immediate post-EVS EEG data to evaluate the performanceof the artifact rejection under the assumption that the difference in brainactivity during and immediately after the stimulation would be minimal.Six epochs were concatenated together to calculate power spectrum in thecleaned and post-EVS EEG data. Then, power differences (Pdiff ) in the543.3. Resultstheta, alpha, beta and gamma bands were calculated for each subject asfollows:Pdiff =1KK∑k=1| P˜k − P¯k | (3.14)where P˜k and P¯k are the power in each frequency band for channel k of thecleaned and post-EVS EEG data, respectively.The performance of the proposed q-IVA method was further investigatedby examining occipital alpha rhythms, as has been previously-used in per-formance evaluation of artifact removal algorithms [187]. Occipital alpharhythms are one of the standard physiological responses that have been in-vestigated in many EEG studies, and are dominant during an eyes-closedresting condition and suppressed when individuals open their eyes [23]. Forthis, we compared the alpha power at the occipital region (channel O1) inthe cleaned EEG between when the subjects eyes were open and closed.Statistical AnalysisWe analyzed whether the group means of the performance evaluation mea-sures were significantly different depending on the artifact rejection meth-ods using an analysis of variance (ANOVA). For the simulation results, theANOVA was performed for each measure with a group factor (MCCA vs.IVA vs. q-IVA) and a within-group factor (EEG channels). For the real datastudy, the ANOVA was performed for the Pdiff in each frequency band witha group factor (MCCA vs. IVA vs. q-IVA) and a within-group factor (10subjects). For each ANOVA test, a Tukey-Kramers test (using the “mult-compare” function in MATLAB) was used for multiple pairwise comparionof the mean values between the three groups.3.3 Results3.3.1 Simulation ResultsFig. 3.4 shows the principal components (PC) from PCA, underlying sourcecomponents (IC) from SOBI, IVA and q-IVA, and canonical variates (CV)from MCCA. The correlation coefficient between the stimulation signal andthe first component from each method was similar 0.976 ± 0.001 (mean± standard deviation) for both PCA and SOBI, 0.969 ± 0.04 for MCCA,0.974 ± 0.03 for IVA, and 0.877 ± 0.04 for q-IVA, indicating a significant553.3. Resultsportion of the artifact was identified in the first components. The correlationcoefficient between the stimulation signal and the second component was0.212 ± 0.005 for PCA, 0.213 ± 0.005 for SOBI, 0.214 ± 0.120 for MCCA,0.292 ± 0.231 for IVA, and 0.263 ± 0.213 for q-IVA, indicating the secondcomponents still had a significant correlation with the stimulation signal.For the third component, the correlation coefficient dropped significantlybelow 0.01 for all methods. Therefore, we removed the first two componentsand reconstructed EEG from the rest components for all the five methods.Figure 3.5: Sample traces (8 channels) of the cleaned EEG data after using differentartifact rejection methods. Note that for PCA and SOBI, artifacts are inconsistentlyremoved across channels, whereas MCCA, IVA and q-IVA removed artifacts robustlyacross all channels and epochs. For the illustration purposes, the first 5-s of the channelsF3 and P7 are shown at the bottom (red: the cleaned EEG; black: the original EEG).Fig. 3.5 shows an example of the cleaned EEG data after using differentartifact rejection methods in one of the iterations. PCA poorly removedthe stimulation artifacts, showing inconsistent performance across channels.It removed artifacts well in a few channels whereas in other channels theartifacts were either insufficiently removed (e.g., F3) or removed with signif-icant amount of non-artifactual signal (e.g., O2). Likewise, SOBI left smallartifactual remnants in some channels (e.g., O1) or removed a substantialamount of non-artifactual signals (e.g., P7). In contrast, MCCA, IVA andq-IVA removed the artifacts robustly across all channels.Fig. 3.6(a) compares the RRMSE values across 27 channels betweenthe different methods. The RRMSE for PCA was highest in all channelsfollowed by SOBI. The JBSS methods, in general, demonstrated much im-proved results in the RRMSE for all channels with the channel-averagedvalue being approximately 5.8 and 1.6 times smaller than PCA and SOBI.Statistical results showed significant differences in RRMSE between the563.3. ResultsFigure 3.6: Comparison of the performance of different artifact rejection methods in thesimulation study (for illustrative purposes, in (a) and (b), the results of MCCA, IVA,and q-IVA in the first panel were taken and magnified in the second panel; the thirdpanel shows the results of ANOVA to test group mean differences; *: P < 0.05, **:P < 0.01, ***: P < 0.001). (a) RRMSE. (b) CC. (c) Power spectrum of the channelO1 of the original data (Ys,EEG), simulation data (Ys), and cleaned data (Y˜s,EEG) afterusing different artifact rejection methods. The bar graphs on the right show the resultsof ANOVA to test group mean differences in Pdev in the theta, alpha, beta, and gammabands.573.3. ResultsJBSS methods; it was smallest for q-IVA (0.219 ± 0.041) followed by MCCA(0.231 ± 0.043) and IVA (0.238±0.057). An interesting finding was that theRRMSE changed across the channels in a similar way for MCCA, IVA andq-IVA. This suggests that the condition in the objective function associatedwith maximizing overall correlation among multiple data sets has strongerpotential for extracting source components than solely restricting sources tobeing uncorrelated or statistically independent.Fig. 3.6(b) shows the correlation coefficients between the original and thecleaned EEG data. It can be seen that MCCA, IVA and q-IVA resulted inCC being close to 1 in almost all channels, indicating superior performanceto PCA and SOBI. Statistical results showed that the CC was significantlyhigher for q-IVA (0.972 ± 0.010), followed by MCCA (0.968 ± 0.012) andIVA (0.966 ± 0.016).The performance of the artifact rejection methods in the frequency do-main is demonstrated in Fig. 3.6(c). The first panel shows an example ofthe power spectrum of the channel O1 of the resting EEG data (black) andafter it was corrupted by the simulated artifacts (red). The power of theartifact was around 50 dB obscuring the alpha peak around at 9 Hz in theoriginal data. For all of the artifact rejection methods, the dominant peakat 9 Hz was again detectable in the cleaned EEG data. However, PCA re-sulted in significantly diminished power in all frequencies suggesting that asignificant amount of EEG signals was removed along with the rejected PCs.SOBI also removed some of the EEG signals as can be seen in the decreasedpower in the frequency range of 6–23 Hz. Although MCCA, IVA and q-IVAalso slightly decreased the power around 9 Hz, the overall power spectrawere much closer to the original EEG compared to PCA and SOBI. In theright side of Fig. 3.6(c), the results of ANOVA comparing group means ofPdev in each frequency band are shown. Q-IVA showed the smallest Pdev inall the frequency bands. In the theta and alpha bands, the Pdev was signifi-cantly lower for IVA than MCCA, while it was significantly greater for IVAthan MCCA in the beta and gamma bands. Overall, the results indicatethat the power spectrum of the EEG data cleaned by q-IVA was closest tothe true value.3.3.2 Real Data ResultsThe simulation results in the previous section demonstrated that the JBSSmethods outperformed PCA and SOBI. In the real data study, the perfor-mance of MCCA, IVA and q-IVA was further investigated.The first panel of Fig. 3.7(a) shows an example of the channel O1 power583.3. ResultsFigure 3.7: Power spectrum of the channel O1 averaged over 10 subjects. (a) Comparisonof the pre-EVS, post-EVS and cleaned EEG data after using MCCA, IVA, and q-IVA. (b)The results of ANOVA to test group mean differences in Pdiff in the theta, alpha, betaand gamma bands (*: P < 0.05, **: P < 0.01, ***: P < 0.001).spectrum of the pre- and post-EVS EEG data averaged over 10 subjects.It can be seen that the power in the beta and gamma bands increased af-ter stimulation, suggesting nonlinear EVS effects on brain activity, as thestimulation was in the theta band. The power spectrum during EVS wasobtained after removing the stimulation artifacts using MCCA, IVA andq-IVA, which is illustrated with the pre- and post-EVS power spectra fromthe second panel. Before artifact rejection, the stimulation artifact was thedominant feature in the spectral power, with the mean power at the stim-ulation frequency (4–8 Hz) of around 55 dB across subjects. The powerspectrum of the cleaned EEG data was comparable to the one in the post-EVS period for all subjects; the increased power in the beta and gammabands resulted from the stimulation effects on brain activity was also de-tectable in the cleaned EEG data. Fig. 3.7(b) shows the Pdiff in the thetaand gamma bands was significantly lower for q-IVA compared to MCCA593.3. Resultsand IVA. The Pdiff of MCCA was significantly higher than q-IVA in thetheta, beta and gamma bands, and higher than IVA in the beta and gammabands. There was no significant difference in the Pdiff in the alpha bandbetween the three methods. Overall, the results suggest that q-IVA followsthe power spectrum of the post-EVS EEG data most similarly.Figure 3.8: The alpha activity at the channel O1 in the eyes-open (EO) and eyes-closed(EC) conditions. (a) Pre-EVS. (b) During EVS. (c) During EVS after applying q-IVA to(b). (d) Comparison of the power spectrum of the pre- and during EVS (before/after theartifact removal using q-IVA).Fig. 3.8(a) shows a spectrogram of the channel O1 when the subjects eyeswere open (EO; 0–60 s) and closed (EC; 60–120 s) in the resting state (i.e.,pre-EVS). The enhanced alpha activity can be seen in the EC condition. Fig.3.8(b) shows the raw EEG signal and its spectrogram in the EO (left) andEC (right) conditions while the EVS was being delivered to the subject. Due603.4. Discussionto the prominent power of the stimulation artifact, the alpha power changebetween the two conditions is not detectable. In contrast, in Fig. 3.8(c),after the artifact was removed using q-IVA, the enhanced alpha activity dueto eye closure can be observed in the EC condition. This result is presentedas power spectra in Fig. 3.8(d). After removing the artifacts with q-IVA,the power in the 4–8 Hz was attenuated close to pre-EVS levels in both EOand EC conditions. The amount of increased power at the alpha peak due tothe eye closure is detectable after applying q-IVA, which is otherwise unableto be detected.3.4 Discussion3.4.1 Advantages of JBSS ApproachesIn this paper, we demonstrated the feasibility of using JBSS methods, andq-IVA in particular, for removing high-amplitude stimulation artifact fromcorrupted EEG data. Several previous studies have suggested using an av-eraged artifact template to remove any stimulation artifacts by subtractingit from the raw EEG signals. The underlying strong assumption of suchan approach is that the stimulation artifacts within each time window isperfectly aligned, since any time lag of the artifact between each windowwould result in significant errors when the averaged template is subtractedfrom the EEG recordings. This assumption is hard to meet in practice,as it is often found that the stimulation artifact appears in the recordedEEG with variable time lags (up to tens of milliseconds). Moreover, theperfect alignment of the stimulation artifacts across windows becomes moredifficult when the frequency of stimulus is relatively low and the length ofeach averaged window becomes wider accordingly. Another potential prob-lem with the artifact template approach is deciding how many windows toaverage. Averaging many windows may increase the chance of having notonly the artifact, but also true underlying brain responses such as entrainedoscillations phase-locked to the stimulation frequency included in the aver-aged template. In order to address this issue, we carried out a regressionon a single-channel and single-trial basis where the stimulus signal and itsquadrature component were used to remove the high-amplitude artifact.Our hypothesis was that the quadrature component would account for anyphase lag between the stimulus and the artifact recorded in EEG, and theregression would preserve information related to brain activity in the errorterms, as the regression matrix only contains stimulus-related information.In the simulation, we demonstrated that IVA was comparable to MCCA613.4. Discussionin the performance of rejecting stimulation artifacts, while q-IVA achievedsignificantly better results in RRMSE, CC and Pdev than both IVA andMCCA. This suggests that removing the high-amplitude artifact first us-ing the quadrature regression might result in IVA being more selective indisentangling artifactual components from EEG records because the statis-tical properties of the remnant artifacts and neural oscillations become moredistinctive.The rationale of using IVA in the removal of the remnant artifacts camefrom the superior performance of JBSS methods in stimulation artifact re-moval compared to PCA and SOBI, which, to the best of our knowledge,we demonstrate here for the first time. Since the objective of PCA is toreduce the dimension of the dataset and explain most of the variability inthe original data, the first principal component accounts for as much of thevariability as possible. For the simultaneous EEG-EVS data, the first prin-cipal component was largely composed of the stimulation artifact (Fig. 3.4).However, it was found that eliminating the first PC alone was insufficientto remove the artifact and accordingly the second PC had to be removed,but this still did not substantially improve the results. The main problemwas that the results varied significantly across channels, as demonstrated inFig. 3.5, which may possibly result from the misalignment of the first PC(i.e., the artifact) in the time domain across the channels due to time lags.ICA has become more widely used than PCA in denoising EEG dataas it is not restricted to the constraint of the spatial orthogonality betweensources and non-brain artifacts are believed to have distinct statistical char-acteristics compared to brain activity. SOBI is an ICA-based algorithm thatfinds the unmixing matrix, W , by minimizing the correlation between onerecovered source at time t and another at time t+ τ by taking into accountthe time delay covariance matrices [26] (technical details can be found in[85]). Considering the algorithm, the poor performance of SOBI shown inthis study implies that the stimulation artifacts and neural activity sharesimilar second-order statistical properties, making separating the two diffi-cult.In contrast to PCA and SOBI, the JBSS methods analyze EEG data inmultiple epochs jointly and take into account the correlation and covarianceof sources in the multiple epochs. This property dramatically improvedthe results for this specific stimulation artifact removal problem since theartifacts appear highly consistent, but not identical, between time windows.Similarly, as in [2], the JBSS methods can be applied to artifact correctionin the simultaneous EEG-fMRI data as the MR artifacts are also consideredquasi-periodic.623.4. Discussion3.4.2 Recommendations and LimitationsThe proposed q-IVA method allows for the investigation of brain stimulationeffects on neural dynamics during stimulation. Although the stimulationtechnique used here is EVS, the developed method could readily be appliedto remove artifacts generated by other NEBS techniques such as tACS. It isalso noteworthy that the tested stimulus signal in this study was a multisine,with a complex wave form compared to a simple pure sine wave that hasbeen used in the majority of tACS studies. Since the period of the thetamultisine was 5 s, we used a 5-s window length for each epoch. In the caseof a single-frequency tACS, the period of the stimulus is much shorter than5 s, which would allow for a greater flexibility in selecting an appropriatewindow size for the same length of EEG.There are several limitations to this study. The results are reportedbased on a case where the theta-multisine was used as the stimulus. Wehave not investigated effects of the number of epochs, epoch length, fre-quency of stimulus, and the number of EEG channels on performance re-sults, which need to be thoroughly investigated in a future study to obtainfurther improvement. In addition, we note that the q-IVA method maynot be appropriate for removal of MR gradient artifacts in simultaneousEEG-fMRI recordings, as the artifacts are a sequence of pulses rather thana continuous sinusoidal signal and accordingly the regression based on aquadrature component may not be applicable. Although we demonstratedsuperior performance of q-IVA compared to the conventional methods, wenote that careful interpretation of the results should be made. After re-moving the artifacts, any changes observed during stimulation compared topre-stimulation period could be true online effects of stimulation or errorsintroduced by the data processing. The interpretation is still not definitiveas the ground truth is unknown. Here we proposed that comparison of thecleaned EEG data with the immediate post-stimulation data can be one wayto minimize erroneous interpretation of online stimulation effects in futureNEBS studies.In summary, we have investigated the performance of various methods toattenuate stimulation artifacts using quantitative measurements in simula-tions. In contrast to the conventional methods such as PCA and SOBI, theJBSS methods (MCCA, IVA and q-IVA) substantially improved the perfor-mance and the proposed method, q-IVA, outperformed all other methods.It was not investigated here whether applying the quadrature regression be-fore PCA and SOBI would improve their denoising performance since ourprimary interest was the JBSS-based methods, MCCA and IVA, and they633.4. Discussiondemonstrated significantly superior performance to PCA and SOBI regard-less of the quadrature regression. When examining the real data, we demon-strated that the q-IVA successfully attenuated the stimulation artifact, en-abling the detection of the stimulation effects that resembled those seen inthe post-stimulation data as well as physiological change by eye closure inthe cleaned EEG data, which would be otherwise completely obscured bythe high-amplitude artifacts. The results of this study suggest that q-IVAis an effective approach for the investigation of neurophysiological onlineeffects of NEBS.64Chapter 4Sparse Discriminant Analysisfor Detection of PathologicalDynamic Features of CorticalPhase Synchronizations inPDIn this chapter, we demonstrate that pathological cortical activities in PDare normalized while EVS is being delivered. First, we introduce novel mul-tisine stimuli that have several advantages over the noisy stimulus used inChapter 2 to investigate online effects of EVS. Next, using the q-IVA denois-ing method introduced in Chapter 3, we remove the stimulation artifacts andinvestigate the changes in EEG during the stimulation. Finally, we providea means towards optimizing EVS by demonstrating that the normalizingeffects of EVS are dependent on the stimulation frequencies.4.1 IntroductionPD, the second most common neurodegenerative disease [335], is character-ized by motor symptoms such as bradykinesia, tremor, rigidity and impairedbalance and gait as well as non-motor complications, resulting primarilyfrom degeneration of dopaminergic neurons in the substantia nigra parscompacta (SNc) [79]. Several electrophysiology studies using local field po-tential (LFP) recordings demonstrated that, in the dopamine-deficient state,the neuronal synchronization in the basal ganglia is exaggerated at frequen-cies in the beta range (13–30 Hz) [51, 100, 219, 275]. These beta oscillationsare also highly synchronized with sensorimotor areas [50, 60, 233, 427] aswell as muscle activity of upper limbs during movement [233]. This exces-sive beta synchronization is considered to be, in part, responsible for theParkinsonian symptoms and thus reducing the abnormal synchronization654.1. Introductionwith deep brain stimulation (DBS) has shown to be an effective therapy.Recent fMRI findings have highlighted that large-scale cortical resting-state functional connectivity (rsFC) is altered in PD, possibly as a resultof BG impairment effects on cortical-BG networks [148]. The striatum, asubcortical region significantly affected with dopamine depletion in PD, hasaltered FC with inferior parietal, temporal, and motor cortices [148], whichsupports that PD-induced connectivity changes can be seen beyond localsubcortical regions. In addition to effects on BG-cortical FC, impairment inthe BG can also alter cortico-cortical connectivity. Diminished interhemi-spheric connectivity in sensorimotor cortical regions [344] and reduced rsFCin widespread regions including inferior frontal, superior parietal, and oc-cipital regions [95] have been shown to be implicated with disease durationand cognitive dysfunctions in PD.Inferring pathological cortico-cortical connectivity in PD solely based onevidence from fMRI alone may not provide a complete picture, as fMRI haslimited temporal resolution. Electrophysiology can provide complementaryinformation as it measures spontaneous synchronous activity of a large pop-ulation of neurons occurring on a millisecond time scale. A simultaneousLFP-electroencephalography (EEG) study reported that the dynamics ofLFP synchrony in the STN is related to the dynamics of cortical synchrony[4], and BG DBS modulates cortical phase coupling measured with the EEG[351, 355].One of the most widely-used method to quantify the coupling betweenoscillatory signals recorded at pairs of electrodes placed on the scalp inEEG is to look at their phase relationships [103, 184]. If cortical activitiesat two different regions are coupled, their phase angle differences tend to beconsistent across time. Phase locking value (PLV) quantifies the strength ofthe phase coupling between two oscillatory signals, bounded between zeroand one indicating a completely random and perfectly coupled relationship,respectively. Interregional phase synchronization has been shown to reflectspecific neural activity coding different cognitive functions [141, 183], motorbehaviours [13] (for a review, see [333]) and pathological brain states [207,359, 394]. However, to date, only a few studies have examined phase-basedrsFC across broad cortical regions and different frequency bands in PD [124,144, 254, 351].Most EEG connectivity studies to date have employed magnitude squaredcoherence. PD subjects exhibit excessive EEG coherence [124, 351], espe-cially in the beta band, in the off-medication condition that is decreased bymedication [124]. For PD subjects on-medication, enhanced coherence in thefrontal regions in the theta (4–6 Hz), beta (12–18 Hz), and gamma (30–45664.1. IntroductionHz) [254] and altered interhemispheric beta coherences in the midtemporaland frontal areas [144] can be observed, indicating the multifarious role ofdopamine in the control of oscillatory activity, in and beyond the BG. How-ever, coherence is different from PLV in that it relies on the assumption oflinearity and stationarity in the signals and is calculated independently foreach frequency, which is then scaled by the amplitudes of the signals. PLV-based connectivity, which do not rely on the strict assumptions underlyingcoherence, might be more suitable for nonlinear and non-stationary dynam-ics of neural oscillations, and sheds a new light on pathophysiological brainnetworks as it has not been explored yet in PD.Recent progress in non-invasive brain stimulation (NIBS) has demon-strated its capability to modulate cortical oscillations [10, 147, 408] andinterregional couplings, indicating its potential applications as an effectivetherapeutic technique for PD. EVS is a NIBS technique that delivers weakcurrent to the mastoid processes and modulates firing rates of vestibularafferents, which then activates various cortical and subcortical regions in-cluding the BG and thalamus [30, 222, 392]. Similar to transcranial elec-trical stimulation (tES), EVS stimuli can take the form of direct current(DC), alternating current (AC) or random noise (RN) and stimulation ef-fects vary according to stimulus types. While DC-EVS perturbs perceptionof orientation and locomotion and has been widely utilized in postural bal-ance control research [362], RN-EVS has demonstrated its efficacy in motorfunctions [205, 278, 432] and modulation of EEG oscillatory rhythms acrossbroad cortical regions in PD [177]. It is conceivable, therefore, that EVSmay be able to modulate cortical couplings, which has not been exploredyet.To establish the potential of EVS as a therapeutic intervention to modu-late cortical couplings in PD, we investigate resting-state cortical couplingsmeasured as PLV that are altered in unmedicated PD patients and nor-malizing effects of EVS. Specifically, we applied three novel EVS stimulibounded into specific frequency bands to PD and healthy subjects and ex-amined whether EVS normalizes both the strength and temporal variationof aberrant couplings in PD and the effects are varying according to thestimulation frequencies.674.2. Materials and MethodsTable 4.1: Demographic and clinical characteristics of the patients with Parkinson’s disease(PD) and healhty controls (HC)PD HCAge (years), mean (SD) 67.3 (6.5) 67.6 (8.9)Gender, n (male/female) 7/9 9/9Disease duration (years), mean (SD) 7.4 (4.3) -UPDRS II, mean (SD) 14.8 (8.1) -UPDRS III, mean (SD) 22.1 (8.9) -Hoehn and Yahr scale, mean (range) 1.3 (1-2) -Levodopa Equivalent Daily Dose (mg), mean (SD) [383] 635.9 (356.4) -UPDRS II: Motor aspects of experience of daily livingUPDRS III: Motor symptoms4.2 Materials and Methods4.2.1 ParticipantsTwenty PD patients and 22 age- and gender-matched healthy controls (HC)participated in this study. Patients with atypical parkinsonism or otherneurological disorders were excluded from the study, and all included PDpatients were classified as having mild to moderate stage PD (Hoehn andYahr Stage 1-2). Four PD and four HC subjects were excluded in the dataanalysis due to severe muscle artifacts in their EEG recordings. Therefore,16 PD (7 males; age: 67.3 ± 6.5 years) and 18 HC (9 males; age: 67.6 ±8.9 years) subjects were included in the analysis (Table 4.1). All subjectsdid not have any reported vestibular or auditory disorders and were right-handed. The study protocol was approved by the Clinical Research EthicsBoard at the University of British Columbia (UBC) and the recruitmentwas conducted at the Pacific Parkinsons Research Centre (PPRC) in UBC.All subjects gave written, informed consent prior to participation.684.2. Materials and MethodsStudy ProtocolAs individuals have inherently subjective perception of EVS, we utilizedsystematic procedures that have been previously used in determining sub-liminal current level [205]. The measured individual threshold level was inthe range of 0.23–1.1 mA. After the threshold was determined, the subjectswere comfortably seated in front of a computer screen and were instructedto focus their gaze on a continuously displayed fixed target while EEG wasbeing recorded. EEG was first recorded without stimulation for 20 s andEVS were then delivered for a fixed duration of 60 s, followed by an EVS-offperiod for 20 s (post stimulation). During the stimulation period, EVS wasapplied at 90% of the individual threshold level.EEG was recorded from the subjects in 4 different conditions: Sham(no stimulation), EVS1, EVS2 and EVS3 (for details, see 2.3 EVS). EEGrecording was first performed in the sham condition and the EVS conditionswere randomly ordered. We allowed a 2-minute break between each condi-tion to prevent any potential post-stimulation effects carried over from theprevious EVS conditions.The HC subjects performed the protocol once, whereas PD subjects per-formed it twice in off-medication (PDMOFF) and on-medication (PDMON)conditions on the same day. The PD subjects stopped taking their normal L-dopa medication at least 12 hours, and any dopamine agonists 18 hours priorto the EEG recording. United Parkinsons Disease Rating Scale (UPDRS)Parts II and III were assessed in the off-medication condition. Immediatelyafter finishing the EEG acquisition, they took their regular dose of L-dopamedication and rested for one hour. After the break, EEG was recorded inthe on-medication condition. While this did not allow for counterbalancingbetween pre-medication and post-medication conditions, it was felt the vari-ability induced by bringing people in on different days would actually be agreater source of variability than the ordering of PDMOFF and PDMON.4.2.2 EVSEVS was delivered through pre-gelled Ag/AgCl electrodes (BIOPAC Sys-tems Inc., CA, USA) placed in bilateral, bipolar fashion over the mastoidprocess behind each ear. NuprepTM skin prep gel was used to clean skin forbetter electrode contact and to reduce resistance during stimulation. Stim-ulation waveforms were generated on a computer using MATLAB (R2018a,MathWorks, MA, USA) and converted to an analog signal using a NI USB-6221 BNC digital acquisition module (National Instruments, TX, USA). The694.2. Materials and MethodsFigure 4.1: The multisine stimuli and the phase locking value (PLV) calculation. (A)Time and frequency plots of the three types of multisine stimulus given at 90% individualthreshold level (EVS1: 4-8 Hz; EVS2: 50-100 Hz; EVS3: 100-150 Hz). (B) Placement of 27EEG electrodes and PLV calculation. The Hilbert transform is applied to the two signalsto extract instantaneous phases. The phase differences calculated at each time point arerepresented as unit vectors in the complex plane and PLV is computed to evaluate thespread of the distribution (Lachaux et al. 1999; Mormann et al. 2000). (C) The procedureto extract PLV time series. For each subject, preprocessing steps were first applied to theraw EEG data in order to remove high-voltage stimulation artifacts as well as cardinalartifacts caused by eye movements (electrooculography (EOG)) or muscle movement. Thecleaned data were bandpass filtered into 4 different frequency bands (theta: 4–8 Hz; alpha:8–13 Hz; beta: 13–30 Hz; gamma: 30–45 Hz) and segmented into epochs. PLV between apair of electrodes in each epoch was computed to generate the time series, and its mean,variability, and sample entropy were calculated. Each subject has a 1 × p vector for themean, variability and sample entropy (p = 1,404 = 351 pairs x 4 frequency bands)analog voltage signals were then passed to a constant current stimulator(DS5, Digitimer, UK), which was connected to the stimulating electrodes.Three multisine signals in different frequency bands (EVS1: 4–8 Hz;EVS2: 50–100 Hz; EVS3: 100–150 Hz) were used (Fig. 4.1A). Multisinesignals are designed to concentrate power at a precise number of frequencieswithin the bandwidth of interest, which is advantageous compared to otherexcitation signals (e.g., a white noise or swept sine) as there is no spectralleakage. Each multisine signals were designed to have the frequencies ofsinusoids (fi) uniformly distributed every 0.2 Hz and the phases (φi) chosento minimize the crest factor using a clipping algorithm [401] in order to704.2. Materials and Methodsgenerate a flat amplitude of the signal and thus improve subjects comfort:x(t, φ) = a ·n∑i=1cos(2pifit+ φi) (4.1)where x(t, φ) is the multisine, a is the amplitude, and fi and φi are thefrequency and phase, and i is the index of each sinusoidal component (e.g.,f1, f2, ..., fn = 4.0, 4.2, ..., 8.0 Hz for EVS1).4.2.3 EEG recordingData were recorded from 27 scalp electrodes using a 64-channel EEG cap(Neuroscan, VA, USA) and a Neuroscan SynAmps2 acquisition system (Neu-roscan, VA, USA) at a sampling rate of 1 kHz. Recording electrodes werepositioned according to the International 10-20 placement standard withone ground and one reference electrode located between Cz and CPz (Fig.4.1B). Impedances were kept below 15 kΩ using Electro-Gel (Electrode-CapInternational, OH, USA). No clipping of EEG was observed during stimu-lation4.2.4 EEG preprocessingThe EEG data were bandpass filtered between 3 and 45 Hz using a two-way finite impulse response (FIR) filter (the eegfilt function in EEGLAB).High-voltage stimulation artifacts during EVS2 and EVS3 were removedusing the digital filters. The artifacts during EVS1 were removed using aquadrature-IVA method [206]. Data were then re-referenced to the averagereference (linked earlobe) and ocular artifacts (EOG) were corrected basedon cross-correlation with the reference EOG channels using the AAR tool-box included in EEGLAB. The cleaned EEG data were bandpass filteredinto four conventional EEG frequency bands [132]: theta (4–8 Hz), alpha(8–13 Hz), beta (13–30 Hz) and gamma (30-45 Hz). The bandpass-filtereddata were then segmented into non-overlapping epochs. Epoch sizes weredetermined such that the epochs include around 4 cycles at a centre fre-quency of the selected bandwidth, resulting in epoch sizes of 600, 400, 200,and 100 ms for the theta, alpha, beta, and gamma bands.4.2.5 Phase Locking Value (PLV)PLV evaluates the spread of the distribution of phase angle differences be-tween pairs of electrodes over time [195, 256] (Fig. 4.1B). The connectivity714.2. Materials and Methodsis measured from this spread such that strongly clustered phase differencesbetween two electrodes result in the PLV value close to one, indicating astrong connectivity between the signals. If there is no phase dependence,PLV value becomes zero.To calculate the PLV, instantaneous phase angles were obtained by ap-plying the Hilbert transformation to the bandpass-filtered data. Then, thePLV between two signals A and B was computed as [54]:PLVA,B =1T∣∣∣∣∣T∑t=1ei(ϕA(t)−ϕB(t))∣∣∣∣∣ (4.2)where T is the number of time points and ϕ(t) is the instantaneous phaseangles of each EEG signal.The PLV was computed for each epoch, resulting in times series of thePLV computed from all pairs of 27 electrodes and the 4 frequency bands(1,404 time series in total). Three temporal features were extracted fromeach PLV time series for further analysis: the mean, variability (standarddeviation), and sample entropy. Sample entropy is a nonlinear measureto quantify the degree of complexity in a time series [315], and has beenapplied to EEG data for clinical application such as classification [52, 193]and epilepsy detection [358]. Tolerance (r) and window length (m) werespecified to be 0.3 and 2, respectively, to compute the sample entropy basedon [198] and characteristics of our data sets.4.2.6 Sparce Discriminant AnalysisLinear discriminant analysis (LDA) is a classical supervised classificationtechnique that finds the most discriminative projections of a N × p data ina p-dimensional space such that the data projected into the low-dimensionalsubspace can be well partitioned into K classes [228]. In biomedical research,it has become an increasingly important topic to perform classification onhigh-dimensional data where the number of variables far exceeds the numberof samples. In such high-dimensional settings, LDA cannot be applied di-rectly because of singularity of the sample covariance matrix. To overcomethis limitation, various regularized versions of LDA have been proposed[326]. Sparse discriminant analysis (SDA) was proposed by Clemmensenand colleagues [70] where an elastic net penalty and an optimal scoringframework are applied to a high-dimensional data to generate a sparse dis-criminant vector. The authors demonstrated that SDA outperforms otherregularized methods such as shrunken centroids regularized discriminant724.3. Resultsanalysis and sparse partial least squares regression. The details of the algo-rithm can be found in [70].Here, we aim to classify the PDMOFF and HC groups in the baselineresting state (i.e., the sham condition) using the PLV features obtainedabove. The three data sets (mean, variability and sample entropy) have thesame high-dimensional settings as each data set has the number of variables(p = 1,404) much greater than the number of samples (i.e., subjects). There-fore, we applied SDA to each data set to infer from the sparse discriminantvectors which combination of the electrode pairs and frequency bands arethe most important features for the classification of the two groups. As in[70], we created the training set consisted of 26 subjects (12 PDMOFF and14 HC) and the test set of 8 subjects (4 PDMOFF and 4 HC subjects) andthe tuning parameters for SDA (i.e., λ and γ for regularization penalties)were chosen using leave-one-out cross-validation (LOOCV) on the trainingdata. The models with the selected parameters were evaluated on the testdata.In the subsequent analyses, we investigated effects of L-dopa medicationon the PLV features by applying the sparse discriminant vectors obtainedfrom the above SDA to the data sets of the PDMON group in the shamcondition. In the same manner, effects of EVS on the PLV features wereevaluated by applying the same sparse discriminant vectors to the data setsin the EVS conditions.4.2.7 Statistical AnalysisOne-way ANOVA was performed to compare the PLV features betweengroups followed by post-hoc Tukey’s honestly significant difference (HSD)test for multiple comparison correction. To evaluate effects of EVS on thePLV features within a group, repeated measures (rm) ANOVA with stimula-tion condition (sham, EVS1, EVS2 and EVS3) as the within-subject factorwas performed followed by post-hoc Tukey’s HSD test for multiple compar-ison correction. The rm ANOVA was performed for online and after-effect,respectively.4.3 Results4.3.1 SDA Classification Results and Selected FeaturesSDA was performed for the mean, variability, and entropy PLV data sets in-dependently to discriminate the PDMOFF and HC groups. Since there are734.3. Resultstwo classes in the data, only one discriminant vector was obtained from eachSDA. For the mean PLV data set, LOOCV on the training data resulted inthe selection of 17 nonzero features (1.2 %) out of total 1,404 features (Fig.4.2A). There were both negative and positive weights for the selected fea-tures in each frequency band. Since the transformed PLV mean was greaterfor the PDMOFF (Fig. 4.3A) than the HC group, the positive weightswere interpreted as cortical couplings exaggerated in the PDMOFF group.35% of the selected features were associated with Cz over a broad frequencybandwidth, and the PDMOFF group had a stronger coupling strength forthe features. In contrast, the features related to C4 had negative weights,indicating that these couplings are attenuated in the PDMOFF group. Inthe gamma band, decreased long-distance connectivity in the left temporalregion (T7-O1 and T7-P8) and increased short-distance connectivity in theparietal region (P3-PO5, P8-P4, and P8-PO6) were found to be related tothe PDMOFF group. The training and test classification accuracy (fractionof correctly classified) were both 100%.For the PLV variability data set, 12 nonzero features (0.85%) were se-lected and the largest number of the selected features was found in thetheta band (Fig. 4.2B), followed by the alpha and gamma bands. Notethat positive weights are associated with the lower connectivity variabilityof the PDMOFF group because the transformed variability is lower for thePDMOFF group (Fig. 4.3A). Decreased variability in the PDMOFF groupwas mostly associated with the frontal electrodes in the theta band and withF3-Cz, C3-Pz and P7-PO6 in the alpha band. The classification accuracyfor the training and test data sets were 100% and 87.5%, respectively.The SDA on the PLV entropy data set selected 17 nonzero features(1.2%) and most of them were long-distance connectivity. Note that posi-tive weights are associated with the connectivity with lower entropy for thePDMOFF group. In the theta and alpha bands, the entropy of the selectedfeatures was lower whereas in the gamma band the entropy was higher forthe PDMOFF group compared to the HC group. In the beta band, thePDMOFF group had a lower entropy for Fz-O2 and higher entropy for Pz-PO6 than the HC group. The training and test classification accuracy were96% and 87.5%, respectively4.3.2 Group Comparison of Baseline PLV FeaturesThe SDA discriminant vectors were applied to the data sets obtained fromthe PDMON group, and the group means of the transformed data are com-pared in Fig. 4.3A. Significant group differences were found for the PLV744.3. ResultsFigure 4.2: Nonzero features selected by sparse discriminant analysis (SDA). SDA wasapplied to the mean, variability and entropy data sets, respectively, to discriminate thePDMOFF and HC group. The nonzero weights in the sparse discriminant vectors arepresented in the scalp maps. (A) Weights for the 17 selected features from the mean PLVdata set. (B) Weights for the 12 selected features from the PLV variability data set. (C)Weights for the 17 selected features from the PLV entropy data set.features (PLV mean: F(2, 47) = 41.68, P < 0.001; PLV variability: F(2,47) = 23.46, P < 0.001; PLV entropy: F(2, 47) = 60.59, P < 0.001). ThePLV mean for the PDMOFF group was significantly higher than the HCgroup (P < 0.001), which was decreased by L-dopa medication (P < 0.001).The PLV variability was significantly lower in the PDMOFF compared tothe HC group (P < 0.001), and the lower variability was associated withhigher UPDRS II scores (i.e., more severe difficulties of daily motor ac-tivities) (r = −0.56, P = 0.025). The medication slightly improved thevariability in the PD subjects but the changes did not reach statistical sig-nificance (P = 0.096). The entropy of the PDMOFF group was lower thanthe HC group (P < 0.001) and the lower entropy was related to a longerdisease duration (r = −0.56, P = 0.038). The medication did not improvethe PLV entropy (P = 0.21).754.3. ResultsFigure 4.3: (A) Group comparison of the discriminant component obtained from theSDA. The discriminant components were obtained by multiplying the discriminant vectorsto the data sets from the sham condition. Bars and error bars indicate group meansand s.e. Significant P -values from one-sample/two-sample t-tests are indicated (***P <0.001). (B) Pearson correlations with clinical scores. The PLV variability and entropy ofthe PDMOFF subjects are significantly correlated with UPDRS2 and disease duration,respectively4.3.3 Online- and after-effects of EVSNext, EVS effects on the PLV features were investigated. Specifically, weexamined whether the effects are dependent on the stimulus types and sus-tained even after the stimulation ceases. Fig. 4.4A–C show changes inthe PLV mean for each group induced by EVS1, EVS2 and EVS3, respec-tively. The PLV mean was significantly modulated in PDMOFF (F(3, 45)= 11.16, P < 0.001) and HC (F(3, 51) = 3.81, P < 0.05) groups duringstimulation. All stimuli decreased the PLV mean in the PDMOFF groupcompared to the sham condition (EVS1: P < 0.001; EVS2: P < 0.01;EVS3: P < 0.01), making it closer to the HC group, and the effects lastedin the post-stimulation period. EVS1 decreased the mean PLV greater thanthe other two stimuli and there was no continuing decrease in the post-stimulation period whereas EVS3 decreased the mean PLV less than EVS1during stimulation and the effect continued in the post-stimulation period.In contrast, we found the opposite EVS effects for the HC group where EVSincreased the PLV mean (EVS2: P < 0.05; EVS3: P < 0.01). No significanteffects of EVS were found in the PDMON group (F(3, 45) = 0.77, P = 0.52).EVS effects on the PLV variability are presented in Fig. 4.5A–C. There764.3. ResultsFigure 4.4: Effects of EVS on the PLV mean. The PLV mean values in the sham conditionare identical to those in Fig. 4.3A. The PLV mean values in the stimulation (60 s) andpost-stimulation period (20 s) were obtained in the same manner by multiplying thediscriminant vector to the corresponding data sets. In each row, from the left, the resultsfor the PDMOFF (blue), PDMON (green), and HC (grey) groups are presented in eachpanel. Significant P -values from the repeated measures ANOVA with post-hoc Tukey’sHSD test are indicated (*P < 0.05; **P < 0.01; ***P < 0.001). (A) EVS1 effects. (B)EVS2 effects. (C) EVS3 effects.were significant online effects of stimulation on the PLV variability in PDMOFF(F(3, 45) = 4.43, P < 0.01) and HC (F(3, 51) = 4.62, P < 0.01) groups.EVS1 and EVS2 were found to have positive effects on the PDMOFF group,774.4. Discussionincreasing the variability during stimulation (EVS1: P < 0.01; EVS2:P < 0.05). Similar to the effects on the PLV mean, EVS1 induced thegreatest increase in the variability during stimulation and the increased valuetends to return to the baseline after the stimulation ceased whereas the ef-fects of EVS2 and EVS3 were less during stimulation but lasted longer thanthat of EVS1. In the HC group, we found decreases in the PLV variabil-ity induced by EVS (EVS1: P < 0.01; EVS2: P < 0.05; EVS3: P < 0.05).EVS1 decreased the variability during the stimulation and the effect lasted inthe post-stimulation period. EVS2 and EVS3 appeared to further decreasethe variability in the post-stimulation period. For the PDMON group, allstimuli increased the PLV variability but the effects did not reach statisticalsignificance (F(3, 45) = 1.13, P = 0.35).Fig. 4.6A–C show EVS effects on the PLV entropy. The PLV entropy wassignificantly modulated in PDMOFF (F(3, 45) = 4.65, P < 0.01), PDMON(F(3, 45) = 3.12, P < 0.05), and HC (F(3, 51) = 4.25, P < 0.01) groupsduring stimulation. We found that all stimuli increased the entropy signif-icantly in the PDMOFF group (EVS1: P < 0.01; EVS2: P < 0.05; EVS3:P < 0.05), bringing it closer to the HC group. The effects were greatestduring stimulation and diminished in the post-stimulation period, and EVS1increased the largest amount of the entropy, followed by EVS2 and EVS3.For the PDMON group, EVS1 (P < 0.05) and EVS2 (P < 0.05) increasedthe entropy significantly. While not statistically significant, increases in theentropy were also found during and post EVS3 compared to the sham con-dition. The PLV entropy of the HC group changed in the opposite directionby EVS compared to the PD groups. Significant decreases in the entropywas observed with all stimuli (EVS1 (P < 0.01), EVS2 (P < 0.05) andEVS3 (P < 0.01)).4.4 DiscussionWe investigated phase-based cortical connectivity in resting EEG in PD. Toour knowledge, this is the first study that examined connectivity dynamics inPD by characterizing temporally fluctuating cortico-cortical couplings overbroad frequency bands. The results from the current study on the time-varying connectivity provide novel insights into altered cortical dynamicsderived from pathological BG changes in PD.784.4. DiscussionFigure 4.5: Effects of EVS on the PLV variability. The PLV variability values in the shamcondition are identical to those in Fig. 4.3A. Descriptions for the arrangement of theplots and statistical significance are same as in the Fig. 4.4. (A) EVS1 effects. (B) EVS2effects. (C) EVS3 effects.4.4.1 Disrupted Cortical Coupling Strength in the MotorRegionsWe found most changes in cortical coupling strength associated with PD(Fig. 4.2A; 11 out of 17) were in key motor and parietal regions, includ-ing over the primary motor cortex (M1), supplementary motor area (SMA),premotor area (PMA), and superior parietal regions, which was in line with794.4. DiscussionFigure 4.6: Effects of EVS on the PLV entropy. The PLV entropy values in the shamcondition are identical to those in Fig. 4.3A. Descriptions for the arrangement of theplots and statistical significance are same as in the Fig. 4.4. (A) EVS1 effects. (B) EVS2effects. (C) EVS3 effects.previous findings [276]. Typically, a common finding of pathological syn-chronization in PD is hypersynchronization of the cortical regions in thebeta range [124, 301, 351]. This appears to be related to exaggerated betasynchronization within the BG and between the BG and motor cortical re-gions [44, 48, 162]. However, growing evidence indicates that PD has morecomplex influences on motor networks beyond excessive beta synchroniza-tion [430, 431]. There is altered cortical oscillatory activity in other bands804.4. Discussionbeside beta [40, 370]. On the other hand, there is substantial agreementthat therapeutic DBS [301] and dopaminergic medication [146, 374, 431]have normalizing effects on rsFC of motor networks in PD. Consistent withthese findings, our results demonstrated that the altered connectivity foundin the PDMOFF group was normalized by both medication and EVS to asimilar extent.4.4.2 Variablity and Entropy of PLV in the Theta BandThe altered variability and entropy of PLV in the PD group were mostlyfound in the theta band (Fig. 4.2B–C), which may reflect abnormalitiesin thalamocortical dynamics. The ventral anterior (VA) and anterior partof ventral lateral (VLa) thalamic nuclei are the major recipients from theglobus pallidus internus (GPi) via pallidothalamic tracts that are cruciallyinvolved in motor disorders such as PD [120]. Simultaneously-recorded LFPin the VA and VLa nuclei and EEG on the scalp from PD subjects demon-strated the highest coherence in the theta band (4–9 Hz), in particular in thefrontal region of both hemispheres [332]. Thalamocortical interaction maythus be a major influence in generation of frontal theta activity in PD, andpossibly also healthy controls, but we typically do not have LFP recordingsfrom healthy subjects. Multimodal functional imaging studies in healthyhuman and animal models suggest that the thalamus is critically involvedin generating and modulating activities in the cortex [153, 182, 185, 343].The enhanced synchronization in the theta band of the thalamus and frontalcortical region may be reflective of pathological changes in PD. Together,we conjecture that the increased mean and reduced variability in theta thatwe observed in PD subjects was a consequence of excessive synchronizationbetween thalamocortical structures.4.4.3 Variability and Entropy of PLV in the Alpha BandThe dominant frequency in the human EEG under rest is in the alpha fre-quency band (8–13 Hz). Alpha oscillations are known to be affected byvisual and auditory stimuli [142] and change during voluntary movement[289]. A large body of evidence has also demonstrated the critical role ofalpha rhythms in attention as well as various cognitive functions [182]. Thedynamic change of alpha activity reflects a variability of states with en-hanced and reduced cortical excitability, facilitating the brain’s responses tosurrounding stimuli [123]. Several studies have shown that brain signal vari-ability/complexity can serve as an important discriminator for clinical com-814.4. Discussionparisons. For example, EEG entropy is related to brain maturity, as adultshave higher entropy compared to children and adolescents [218]. Higherentropy is also correlated with better performance on a working memorytask [243]. Schlee and colleagues found reduced variability of alpha activityduring rest over the temporal cortex for subjects with tinnitus compared tocontrols [337]. Similarly, the reduced variability and complexity of the cor-tical couplings of the PD groups we observed may be related to diminishedmotor and cognitive adaptability, as executive cognitive functions such as setshifting, divided or alternating attention and dual tasking (e.g., combiningwalking with another task) are impaired in PD [1, 258, 415]. Although themechanisms responsible for these symptoms have not been fully accountedfor, dopaminergic depletion in the striatum disrupts the parallel organizationof cortico-striatal circuits, resulting in more widespread instead of domain-specific involvement of striatal activity and loss of the normally segregatedcircuits [31, 58, 263]. Our results together with the close relationships be-tween cortico-striatal circuits and cortical alpha oscillations [202, 354] war-rant future studies to further elucidate the functional implications of theimpaired alpha dynamic couplings we have demonstrated here.4.4.4 PLV sample entropy is higher in the long-rangegamma activity in PDWe found that the connectivity in the gamma band was more irregular inthe PD group than the HC group (Fig. 4.2C). The binding of cortical re-gions together via synchronization of gamma oscillations between neuronalpopulations, is implicated in numerous cognitive processes [116, 356]. Involuntary movement, for example, synchronization of cortical gamma oscil-lations prior to movement onset has been described as representing activeinformation processing [292, 328] and considered to serve as a prokineticsignal [44]. Abnormal gamma oscillations in the motor cortex in PD havebeen reported [219, 270]. However, resting-state gamma oscillations andconnectivity in PD remain largely unknown. The mechanism underlyinggeneration of the gamma oscillations are known to be critically involvedwith excitatory postsynaptic potentials (EPSPs) of gamma-aminobutyricacid (GABA)ergic interneurons and their intact function of fast-spiking[117, 138, 410]. Thus, alterations in function of GABAergic interneuronscould be inferenced from gamma-band oscillations at the macroscopic level.The fast-spiking interneurons are modulated by neurotransmitters includ-ing acetylcholine [116, 376, 385] and serotonin [116, 309], and there is ro-bust evidence demonstrating deficits in the cholinergic and serotoninergic824.4. Discussionsystems in PD contributing to various aspects of parkinsonian pathophys-iology including motor symptoms, gait dysfunction, cognitive decline, au-tonomic dysfunction (for review, see [288]). Therefore, it is likely that thedisrupted neurotransmitter systems in PD cause alterations in the activitiesof fast-spiking interneurons, subsequently resulting in pathological corticalcouplings in the gamma band in PD.4.4.5 Normalizing Effects of EVS and PotentialMechanismsIn this study, we demonstrated that EVS normalizes the mean, variabilityand entropy of PLV in PD subjects during stimulation and the extent andduration of the effects were dependent on the stimulation frequencies (Fig.4.4–4.6). Modulatory effects of EVS on the cortical oscillatory activity werereported in prior EEG studies that noisy stimulus (pink noise in 0.1–10 Hz)decreased gamma oscillatory activity in the lateral regions and increased thebeta and gamma activity in the frontal region [177], and altered interhemi-spheric coherence [204]. To our knowledge, effects of high-frequency EVS (>50 Hz) on cortical activity have not been explored yet in humans and the re-sults presented in this study provide valuable information on how the effectswould differ from low-frequency EVS that has been used in prior behaviourand neuroimaging studies. We found two characteristics of effects inducedby EVS2 and EVS3 on PLV. First, their effects were similar to EVS1 inthe sense that the direction of changes (i.e., increase or decrease in the PLVfeatures) was the same. We did not find a frequency specific increase or de-crease in the PLV value in both the PD and HC groups. Second, the extentof changes was less compared to EVS1 during the high-frequency stimula-tion but lasted longer in the post-stimulation period. This was observed inthe PDMOFF group for all the PLV measures and in the HC group for thevariability and entropy. For the PDMON group, the EVS effects were lesssignificant, indicating the processing of vestibular inputs in the thalamusand BG [221, 368, 421] is dependent on the dopaminergic level of the BG.Modulating of firing rates of vestibular afferents by externally appliedelectrical current will alter directly the vestibular nuclei activities in thebrain stem, and eventually multiple cortical areas through the thalamocor-tical vestibular system. Thus, understanding vestibular information pro-cessing regarding varying frequency contents at the vestibular nuclei andthalamus is critical to comprehend above findings. A prior study that exam-ined spiking rates of the guinea pig medial vestibular nuclei (MVN) reportedthat two types of neurons having different characteristics of afterpotentials834.4. Discussionresponded to current inputs differentially according to the frequency con-tent (1–30 Hz) [316]. It was shown that spontaneous firing rates of type Aneurons was well modulated by only low-frequency (< 10 Hz) current in-puts and the spiking rates becomes irrelevant to the current input at highfrequencies whereas type B neurons tended to fire in synchrony better whenthe stimulation frequency was higher, which demonstrates existence of sig-nal transformation at the vestibular nuclei level to a certain extent in thattype A neurons act like a low-pass filter [94, 316] whereas type B neuronsact as signal detectors with greater sensitivity to external stimuli at highfrequencies.Considering functional roles of the thalamic nuclei playing integrativeand modulatory roles in sensorimotor processing [389], it is likely that fur-ther transformation of the modified signal transmitted from the vestibularnuclei occurs in the thalamus. The VA, VL, ventral posterior lateral (VPL),ventral posterior medial (VPM), intraminar nuclei and geniculate bodies ofthe thalamus receive primary afferents from the vestibular nuclei and playa critical role in processing vestibular information [30, 55, 247, 366, 421].These thalamic nuclei also receive a range of different afferents from pe-ripheral sensory, subcortical, and cortical regions, and process the differenttypes of information before sending the refined signals to the cortex. Thismay also explain the interaction between EVS and L-dopa medication asobserved in the PDMOFF and PDMON groups as the thalamic nuclei pro-cessing vestibular information would be receiving differential inputs from theBG according to dopaminergic state. Together, unlike transcranial electri-cal or magnetic stimulation that directly target cortical regions of interest,influences of EVS on cortical activities are much more indirect. Our resultssuggest that although the frequency contents of current input to the periph-eral vestibular nerve vary considerably, alterations of the refined higher-levelmultisensory information transmitted from the thalamic nuclei to the broadcortical regions may be relatively consistent.4.4.6 LimitationsThe post-stimulation effects were evaluated for the first 20 seconds only afterstimulation ceased and there may be potential confounding effects if the aftereffects persist much longer. After effects of EVS on cortical activation havenot been fully investigated yet. Delayed responses in the beta and gammapower in frontal regions was reported to appear 20–25 s after 72-s EVS,but lasted only for several seconds. Based on prior studies reporting aftereffects of invasive [429] and non-invasive stimulation [372] and the short844.4. Discussionduration of weak current EVS used here, we concluded that the break timeand randomly-ordered trials were sufficient to avoid confounding effects.We note that the PLV may be affected by volume conduction as in thecase of EEG data several electrodes can simultaneously pick up activitiesfrom the same underlying sources. The propagation of the source signalscan be assumed to be instantaneous due to the low capacitance of the skintissues and the small distance that the currents have to travel [266, 364], andthus PLV at zero-phase differences are susceptible to the volume conduction[165]. Several alternative methods such as the phase lag index [364] andspatial filtering have been proposed to address the issue [377].In conclusion, in this resting-state EEG study, we demonstrated thatconnectivity strengths in the sensorimotor region, and variability and com-plexity of the time-varying cortico-cortical connectivity are affected in PD,and improved by subthreshold EVS. Furthermore, the magnitude and du-ration of the improvement was found to vary depending on the stimulationfrequency and the subjects’ dopaminergic status. The findings from thecurrent study provide valuable information that thalamic functions of in-tegrating subcortical afferent inputs and thalamocortical projections to thecortex play a critical role in the mechanism of EVS effects, and warrantfurther investigation of EVS as a potential therapy in PD.85Chapter 5Discriminant CorrelationApproach to JointEstimation of Maximal EVSEffects on Motor Behaviourand Cortical BetaOscillations in PDThe previous chapter demonstrated that pathological cortical couplings inPD can be normalized by multisine EVS. Moving forward, in this chapter weinvestigated whether the multisine stimulus can also improve motor func-tion in PD. Specifically, we focused to answer the fundamental questions1) whether the EVS improves motor task performance in PD, 2) whetherthe EVS modulates movement-related cortical oscillations, and 3) how thechanges in upstream cortical oscillations and downstream movements arerelated.5.1 IntroductionLoss of dopaminergic neurons in the substantia nigra pars compacta (SNc)in PD gives rise to motor and functional changes in the basal ganglia (BG)-thalamo-cortical networks, affecting motor planning and control [83]. Somemotor symptoms may arise because the functional networks in PD are stuckin a fixed state, leaving them in a characteristic exaggerated state of rhythmsresonating in the beta range (13–30 Hz) [100, 210, 420]. Several studies havereported strong beta power in local field potentials (LFPs) recorded from theBG and cortex of PD patients associated with their poverty of movement[420]. PD therapies that restore movement deficits suppress these patho-logically exaggerated, poorly-modulated beta oscillations. For example, ad-865.1. Introductionministration of levodopa medication [50, 191] and DBS [49, 429] attenuatehighly synchronous beta oscillations in the subthalamic nucleus (STN).Normal suppression of beta oscillations in the sensorimotor cortex be-fore and during voluntary movement have long been investigated with EEGand MEG [161] (for review see [175]). This response, event-related desyn-chronization (ERD), has well-characterized temporal features: beta powerstarts to decrease just before movement onset, with sustained suppressionduring movement execution, and returns back to baseline often followed bypost-movement beta rebound (PMBR), predominantly over primary sen-sorimotor regions [175, 290, 291]. PD patients demonstrate distinct ERDalterations, such as delayed onset [82], diminished ERD and PMBR [146],and different topographical patterns [146, 259] compared to controls.Motivated by growing evidence of the functional role of synchronizedneural oscillations in motor deficits and success of DBS in alleviating symp-toms, recent studies have explored the use of non-invasive brain stimulation(NIBS) techniques as a potential therapeutic intervention for PD. Electri-cal vestibular stimulation (EVS) is a NIBS technique that applies electricalcurrent over the mastoid processes to stimulate vestibular afferents that canactivate numerous downstream cortical and subcortical regions [30]. Severalstudies have demonstrated beneficial effects of EVS on motor symptoms inPD [205, 278, 432] and modulatory effects on cortical oscillations [177], buta joint study linking changes in neural oscillations and behaviour inducedby EVS is lacking.Here, using EVS and simultaneously recorded EEG, we investigate mod-ulatory effects of high-frequency EVS on movement-related beta oscillationsand resultant changes in the motor behaviour of PD patients and healthycontrols. Removing the stimulation artifacts in the EEG, online stimu-lation effects were investigated in this study without having to examinesubtle post-stimulation remnants of influence as done in majority of NIBSstudies. Three key findings are reported here. First, EVS augments betaERD over the left motor region before right-handed movement onset andincreases synchronization during motor execution, returning to baseline inbroad frontal and medial parietal regions, resulting in improved motor taskperformance. Second, EVS not only modulates the magnitude of beta ERDbut also influences its timing, resulting in an earlier onset of the ERD peak,and a faster recovery to baseline, suggesting increased fluidity of the motornetwork. Third, these stimulation effects were dependent on behaviouralcontext as the beta oscillations were not significantly altered by EVS duringrest. From the findings, we conjecture that strong vestibular inputs inter-act with movement-related signals (likely in the thalamus) as part of the875.2. Materials and Methods“motor integrative” hypothesis, making it easier to dynamically modulatemotor systems.5.2 Materials and Methods5.2.1 SubjectsThe subjects and study protocol are same as those in Chapter 4. Twenty PDand 22 age-matched healthy control (HC) subjects participated in this study.The PD subjects were classified as having mild to moderate stage PD (Hoehnand Yahr Stage 1-2) without atypical Parkinsonism or other neurologicaldisorders. We excluded four PD and four HC subjects in the data analysisdue to severe muscle artifacts irrelevant to the task such as excessive facialmuscle activity and coughing (we note that clinical characteristics such astremor and bradykinesia scores between included and excluded PD subjectswere not significantly different). Therefore, sixteen PD (7 males, age 67.3± 6.5 years) and eighteen HC (9 males, age 67.6 ± 8.9 years) subjects wereincluded in the analysis (Table 5.1). All subjects did not have any reportedvestibular or auditory disorders and were right-handed.The study protocol was approved by the Clinical Research Ethics Boardat the University of British Columbia (UBC) and the recruitment was con-ducted at the Pacific Parkinsons Research Centre (PPRC). All subjects gavewritten, informed consent prior to participation.5.2.2 EVSEVS was delivered in bilateral, bipolar fashion through pre-gelled Ag/AgClelectrodes (BIOPAC Systems Inc., USA) placed over the mastoid processbehind each ear using a constant current stimulator DS5 (Digitimer, UK).Two multisine signals were used for EVS (EVS1: 100–150 Hz; EVS2: 50–100 Hz). Each signal had the frequencies of sinusoids (fi) uniformly dis-tributed every 0.2 Hz, with the phases (φi) chosen to minimize the crestfactor using a clipping algorithm [401] in order to improve subject comfort.We utilized systematic procedures previously used to determine individualthreshold level [205], and the stimulus was applied at an imperceptible level(90% of sensory threshold) to avoid effects of placebo, general arousal and/orvoluntary selective attention.885.2. Materials and MethodsTable 5.1: Demographic and clinical characteristics of the patients with Parkinsons disease(PD) and healthy controls (HC)PD HCAge (years), mean (SD) 67.3 (6.5) 67.6 (8.9)Gender, n (male/female) 7/9 9/9Disease duration (years), mean (SD) 7.4 (4.3) -UPDRS II, mean (SD) 14.8 (8.1) -UPDRS III, mean (SD) 22.1 (8.9) -Hoehn and Yahr scale, mean (range) 1.3 (1-2) -Levodopa Equivalent Daily Dose (mg), mean (SD) [383] 635.9 (356.4) -UPDRS II: Motor aspects of experience of daily livingUPDRS III: Motor symptoms5.2.3 Study protocolSubjects were comfortably seated in front of a computer screen and in-structed to focus their gaze on a continuously-displayed fixed target for 60s. Then, a written instruction was given to press a key on the keyboardto start the motor task. Subjects were then instructed to respond to a vi-sual cue (“Go”) as fast as possible by squeezing a rubber bulb (Fig. 5.1).This motor task was adapted as it provides more descriptive behaviour mea-sures than button-press tasks and prior studies on hypokinesia demonstratedmotor control abnormalities of the PD subjects via their exerting pressureduring a similar motor task as ours [236]. There were 10 trials in each stimu-lation condition and 12 different stimulation conditions (including sham (nostimulation) in total). The number of trials were selected such that the PDsubjects can complete the entire study protocol without excessive tiredness(particularly in off-medication condition) and significant differences in taskperformance between conditions can still be detected. The order of the stim-ulation conditions was randomized across the subjects and we located thesham condition as far as possible from the EVS conditions by conducting itbefore any EVS conditions in order to avoid any potential effects carried overfrom EVS. We allowed a 2-min break between each condition to prevent any895.2. Materials and Methodsconfounding post-stimulation effects. In this work, we report results fromsham, EVS1 and EVS2 conditions for which we could remove high-voltagestimulation artifacts and analyze the EEG data jointly with the behaviourdata.The PD subjects stopped taking their normal levodopa medication atleast 12 hours, and any dopamine agonists 18 hours prior to the experiment(off-medication; PDMOFF). Unified Parkinson’s Disease Rating Scale (UP-DRS) Parts II and III were assessed in off-medication condition. After thefirst session, they took their regular dose of levodopa medication and restedfor an hour before beginning the second session (on-medication; PDMON).There was one session for the HC subjects.5.2.4 EEG Recordings and PreprocessingEEG was recorded from 27 scalp electrodes with sampling rate of 1 kHzusing Neuroscan SynAmps2 EEG acquisition system (Neuroscan, USA) anda standard electrode cap (64-channels Quik-Cap, Neuroscan, USA). EEGelectrodes were positioned according to the international 10-20 placementstandard with one ground and one reference electrode located between Czand CPz. The electrodes were attached using Electro-Gel (Electrode-CapInternational, USA) and impedances were kept below 15 kΩ. For prepro-cessing, the EEG data were bandpass filtered between 3 and 55 Hz using atwo-way finite impulse response (FIR) filter. The stimulation artifact couldbe removed using the digital filter as they were in the high-frequency range.Data were then re-referenced to the average reference (linked earlobe), andeye blinks, eye movements and muscle activities were removed using inde-pendent component analysis (ICA) available in EEGLAB.5.2.5 Data analysisBehaviour data analysisWe defined five landmarks in the water pressure recordings from the squeez-able pressure-sensor bulb: Pmax (peak grip pressure), t1 (time when thevisual cue was presented), t2 (time when the pressure started to exceed0.05), t3 (time of Pmax), t4 (time when the pressure returns to 0.05 afterPmax). They were used to extract six behavioural indices (Fig. 5.1).905.2. Materials and MethodsFigure 5.1: Schematic of study protocol and motor task performance. (A) Each subjectperformed the study protocol in sham, EVS1 and EVS2 conditions while EEG was con-tinuously being recorded. In each condition, the subjects were comfortably seated andfocused their gaze on a fixed target presented on a computer screen for 60 s (Rest). Afterpressing any key on a keyboard, they started a motor task consisting of 10 trials (Task).Each trial started with a hold phase in which a fixation cross was presented at the centerfor a randomized duration that ranged from 1000 to 2000 ms (N (1500, 500)). Then, avisual cue (“Go”) appeared for 500 ms followed by 1000-ms white blank screen. The sub-jects were instructed to squeeze a rubber bulb as fast as they could to respond to the visualcue. After finishing the motor task, they took a 120-s break. (B) Water pressure recordedfrom the squeezable pressure sensor bulb is plotted along the time in the x-axis. For eachtrial, a peak grip pressure (Pmax) and four time points (t1: the visual cue; t2 movementonset; t3 peak grip pressure; t4: movement termination) were extracted to compute sixbehaviour measures shown in the tableERD AnalysisWith the visual cue onset as the reference time (t = 0 ms), the EEG datawere epoched from -1000 to 1500 ms and then wavelet transformed using915.2. Materials and Methodscomplex Morlet wavelets (center frequency, Ωc = 1; bandwidth parameter,Ωb = 2; 30 frequencies logarithmically distributed from 7 to 50 Hz). Themean beta (13–30 Hz) power in the baseline interval (from -600 to 0 ms),Pbase, was calculated as a reference value to evaluate ERD. Similar to theapproach adopted in [99], we selected two distinct time windows to investi-gate the ERD at different movement phases based on 10th percentile of thereaction time of all subjects: 0–400 ms for motor preparation and 400–1000ms for motor execution. In order to determine if our results were sensitive tothe cut-off times, we repeated the analysis with 0–400 ms and 560–1000 mswindows (based on the 10th and 90th percentile respectively) and the resultswere unchanged (not shown). For each channel, beta ERD was calculatedasERD =1NN∑i=110 log10PiPbase,i(5.1)where P is the mean beta power in the time window, i is the trial index,and N is the total number of trials (N = 10).Joint analysis of behaviour and ERD dataIn order to investigate maximal EVS effects across all subjects with regard tothe task performance and the beta ERD, we used a feature fusion method,discriminant correlation analysis (DCA), that incorporates discriminationacross classes into a canonical correlation analysis (CCA)-based algorithm.The basic concept of DCA is that it searches for transformation weights (WXand WY ) to project original data sets (X and Y ) into a space where the newprojected data (X ′ and Y ′) are correlated with each other and separationof different classes is achieved (for detailed description of the algorithm, see[137]). We created one data set (X) by concatenating the beta ERD fromall subjects during sham (class 1) and EVS (class 2). Another data set(Y ) was created by concatenating the behaviour indices (the peak time wasexcluded due to its collinearity with the squeeze time and reaction time).DCA were performed four times as there were two EVS conditions and twomovement phases. The transformation weights and correlations between X ′and Y ′ were examined to investigate EVS effects on the beta ERD and taskperformance as well as their interrelationship.925.3. ResultsStatistical analysisFor each behaviour index, between-group task performance in the sham con-dition was compared using ANOVA with the results from 10 trials as theresponse variable and subjects and groups as the random and fixed factors.For EVS effects on task performance, ANOVA was performed for each be-haviour index with the results from 10 trials as the response variable andsubjects and EVS conditions as the random and fixed factors. The P valueswere corrected for multiple comparison using Tukey’s honestly significantdifference (HSD) test (multcompare.m in MATLAB). Students paired t-testwere used for the DCA results. For investigation of potential accumulatedand/or learning effects for the repeated stimulation and task, repeated mea-sures ANOVA was performed for each behaviour measure using the taskperformance across the 11 stimulation conditions as the dependent variable,time as the within-subject factor, and group (i.e., PDMOFF, PDMON, HC)as the between-subject factor.5.3 Results5.3.1 Task Performance in the Sham ConditionSignificant group differences in task performance were found in several be-haviour indices (grip strength: F(2, 497) = 8.87, P < 0.001; squeeze ve-locity: F(2, 497) = 3.42, P < 0.05; movement time: F(2, 497) = 4.74,P < 0.01; squeeze time: F(2, 497) = 10.45, P < 0.001; reaction time:F(2, 497) = 12.15, P < 0.001; peak time: F(2, 497) = 11.28, P < 0.001).The PDMOFF group showed a significantly higher peak grip pressure com-pared to the PDMON (P < 0.001) and HC (P < 0.01) groups (Fig. 5.2),and a higher squeeze velocity than the PDMON group (P < 0.05). Boththe movement time and squeeze time of the PDMOFF group were longercompared to the HC group whereas only the squeeze time was significantlylonger than the PDMON (P < 0.001) group. The reaction time and peaktime of the PDMON group were significantly shorter than the PDMOFFand HC groups.5.3.2 EVS Effects on the Task PerformanceThere were significant effects of stimulation on the behaviour indices (Table5.2) with greater effects in the PDMOFF group (Fig. 5.3). Compared to thesham condition, EVS1 significantly decreased the peak grip pressure (P <0.05) in the PDMOFF group, and squeeze time, reaction time and peak time935.3. ResultsFigure 5.2: Motor task performance in sham condition. (A) Comparison of the taskperformance in the sham condition between the PDMOFF, PDMON and HC groups.Significant P values from ANOVAs are indicated (*P < 0.05; **P < 0.01; ***P < 0.001).Error bars indicate SEM.945.3. Resultsin the PDMOFF and HC groups. For the PDMON group, the movementtime was significantly reduced (P < 0.01). EVS2 changed the squeeze timeand peak time in the PDMOFF and HC groups while no significant effectswere found in the PDMON group (Fig. 5.3).In order to determine if there is any interactino between EVS and L-dopamedication for the PD subjects, we carried out two-way repeated measuresANOVA with EVS and medication as within-subject factors and the sixbehaviour indices as dependent variables. The results showed a significantinteraction between the EVS and medication (Wilks’ Lambda = 0.487, F(10,52) = 2.251, P = 0.029).Table 5.2: ANOVA results on the effects of stimulation condition on the behaviour indicesdf Peak grippressureSqueezevelocityMovementtimeSqueezetimeReactiontimePeaktimePDMOFF 2:477 F = 4.19(*P < 0.05)F = 2.73(P = 0.066)F = 7.77(***P < 0.001)F = 12.32(***P < 0.001)F = 3.9(*P < 0.05)F = 18.25(***P < 0.001)PDMON 2:477 F = 1.08(*P < 0.05)F = 1.42(P = 0.24)F = 4.37(*P < 0.05)F = 2.53(P = 0.08)F = 0.3(P = 0.74)F = 0.74(P = 0.48)HC 2:537 F = 0.47(P = 0.63)F = 1.29(P = 0.28)F = 3.6(*P < 0.05)F = 3.9(*P < 0.05)F = 29.04(***P < 0.001)F = 29.13(***P < 0.001)5.3.3 EVS Effects during Motor PreparationCompared to the sham condition, the mean beta ERD during motor prepa-ration was significantly lower during EVS1 (Fig. 5.4A), demonstrating thatEVS1 augmented the ERD in the brain regions indicated in the weight(WX). For better visualization of their spatial locations, WX is presentedon a scalp map (Fig. 5.4B). The largest positive weights were found at C3and CP5, indicating EVS1 augmented the ERD primarily in the left mo-tor regions. During EVS1, the behavioural index was lower compared tothe sham condition (Fig. 5.4A) and was found to be associated more withmovement time, squeeze time, and reaction time as indicated in WY in Fig.5.4B, corresponding to the behavioural results that EVS1 had greater effectson these behaviour indices than the peak grip pressure and squeeze velocity.To infer whether the changes in the ERD during EVS1 were positivelyor negatively related to task performance, we examined its correlation withthe behaviour index and found a positive correlation (r = 0.23, P = 0.019).Given that the lower behaviour index is associated with shorter movementtime, squeeze time and reaction time, a more negative ERD is thus relatedto better task performance. Therefore, the results that EVS1 augments the955.3. ResultsFigure 5.3: EVS Effects on the motor task performance. Experiment conditions areindicated by the light color (sham), dark color (EVS1), and hatch pattern (EVS2). (A)the PDMOFF group (blue). (B) the PDMON group (green). (C) the HC group (grey).Significant P values from ANOVAs are indicated (*P < 0.05; **P < 0.01; ***P < 0.001).Error bars indicate SEM.ERD and decreases the behaviour index can be interpreted as facilitatingmotor function.In post-hoc analyses of the DCA results, we found that the ERD wassignificantly augmented by EVS1 in the PDMOFF (P < 0.05) and HC(P < 0.001) groups (Fig. 5.4C). Particularly, in the PDMOFF group, thedegree of the augmented ERD by EVS1 was positively correlated with dis-ease severity (r = 0.59, p = 0.015; Fig. 5.4D). For the behaviour index,significant improvement by EVS 1 was observed for all groups (PDMOFF:P < 0.001; PDMON: P < 0.01; HC: P < 0.05). As there was no correlationbetween the ERD and behaviour index (r = −0.075, P = 0.46) from DCAresults of the EVS2 condition, we did not draw inferences about the EVS2results.965.3. ResultsFigure 5.4: DCA results demonstrating EVS1 effects on the beta ERD and task perfor-mance during motor preparation. (A) The beta ERD was significantly augmented andthe behaviour index was significantly decreased by EVS1 compared to the sham condition(***P < 0.001; paired t-tests). (B) The weight (WX) associated with the beta ERD ispresented on the scalp map (left), demonstrating that the augmentation of the beta ERDoccurred predominantly in the left motor region. The weight (WY ) for the behaviourindex presented in a bar graph (right) shows that EVS1 decreased the movement time,squeeze time, and reaction time more than the other behavioural measures. (C) Post-hocanalyses with Wilcoxon signed-rank tests revealed EVS1 decreased the beta ERD in thePDMOFF and HC groups and the behaviour index in all groups compared to the shamcondition (*P < 0.05; **P < 0.01; ***P < 0.001). Light and dark colours representthe sham and EVS1 conditions, respectively. (D) Correlation between the degree of aug-mented beta ERD by EVS1 with UPDRS Part III scores. (E) Temporal evolution of thebeta ERD averaged across the subjects in each group (lines: mean; shaded area: SEMof leave-one-out cross-validation). Compared to the sham condition, the suppression ofthe beta power was greater and the timing of the negative peak was earlier in the EVS1condition.5.3.4 EVS Effects during Motor ExecutionThe beta ERD during motor execution was higher during EVS1 comparedto the sham condition (Fig. 5.5A), and the difference was most prominentin the frontal and medial parietal regions (Fig. 5.5B). A negative correlation975.3. Resultswas found between the ERD and behaviour index (r = −0.39, P < 0.001),indicating that a better task performance related to a higher ERD value. Inpost-hoc analyses, considerable increases in the ERD and behaviour indexby EVS1 were found in the PDMOFF (P < 0.01) and HC (P < 0.001)groups (Fig. 5.5C).Figure 5.5: DCA results demonstrating EVS1 effects on the beta ERD and task perfor-mance during motor execution. (A) The beta ERD value was significantly increased andthe behaviour index was significantly decreased by EVS1 compared to the sham condi-tion (***P < 0.001; paired t-tests). (B) The weight (WX) associated with the beta ERDis presented on the scalp map (left), demonstrating the increase occurred in the frontaland medial parietal regions. (C) Post-hoc analyses with Wilcoxon signed-rank tests re-vealed EVS1 increased the beta power during motor execution in the PDMOFF and HCgroups and the behaviour index in all groups compared to the sham condition (*P < 0.05;**P < 0.01; ***P < 0.001). Light and dark colours represent the sham and EVS1 condi-tions, respectively. (D) Temporal evolution of the beta ERD averaged across the subjectsin each group (lines: mean; shaded area: SEM of leave-one-out cross-validation). Com-pared to the sham condition, the rate of the beta power returning to the baseline (t > 400ms) is faster in the EVS1 condition.In the EVS2 condition, similar to the aforementioned results, positiveweights were found in the left frontal and right central regions with someextension to the medial parietal region (Fig. 5.6B), and the ERD over985.3. Resultsthese regions was higher during EVS2 compared to the sham condition (Fig.5.6A). A negative correlation was found between the ERD and behaviourindex (r = −0.36, P < 0.001). In post-hoc analyses, significant increases inthe ERD by EVS2 was observed for all three groups (PDMOFF: P < 0.05;PDMON: P < 0.01; HC: P < 0.001 in Fig. 5.6C). The behaviour index wasreduced by EVS2 in the PDMOFF (P < 0.001) and PDMON (P < 0.01)groups.Figure 5.6: DCA results demonstrating EVS2 effects on the beta ERD and task perfor-mance during motor execution. (A) The beta ERD value was significantly increased andthe behaviour index was significantly decreased by EVS2 compared to the sham condition(***P < 0.001; paired t-tests). (B) The weight (WX) associated with the beta ERD ispresented on the scalp map (left), demonstrating the increase occurred in the frontal andmedial parietal regions. The weight (WY ) for the behaviour index presented in a bar graph(right) shows that EVS2 decreased the peak grip pressure, squeeze time, and reaction timemore than the other behavioural measures. (C) Post-hoc analyses with Wilcoxon signed-rank tests revealed EVS2 increased the beta power during motor execution in all groupsand the behaviour index in the PDMOFF and PDMON groups compared to the shamcondition (*P < 0.05; **P < 0.01; ***P < 0.001). Light and dark colours represent thesham and EVS2 conditions, respectively. (D) Temporal evolution of the beta ERD aver-aged across the subjects in each group (lines: mean; shaded area: SEM of leave-one-outcross-validation). Compared to the sham condition, the rate of the beta power returningto the baseline (t > 400 ms) is faster in the EVS2 condition.995.4. Discussion5.3.5 EVS Effects on Temporal Patterns of the Beta ERDSince EVS had opposite effects on the beta ERD when we divided the motortask into discrete “preparation” and “movement” phases, we probed theorigin of the differential effects by examining the entire time courses of theERD (Fig. 5.4E) (note that the ERD lines represent time courses of the betapower over the brain regions indicated in the DCA weights (WX) shown inFig. 5.4B). Compared to the sham condition, distinct features in the EVS1condition were found at the time when the curve reaches the minimum:suppression of the beta power was greater and the timing of the negativepeak was earlier.For movement execution, differences between the sham and EVS1 con-ditions were observed after the negative peak appeared (Fig. 5.5D). Thebeta power started to return to baseline around 400 ms after the visualcue onset, and the recovery rate was faster during EVS1. In addition, to-wards the termination of the movement, beta power was higher, suggestinga greater PMBR. Similar results were found in the EVS2 condition (Fig.5.6D). Based on these observations, we conclude that the overall increasedERD by EVS during motor execution is ascribed to this temporal feature ofthe beta power recovery rather than the magnitude of the ERD itself beingsmall.5.3.6 EVS Effects on the Beta Oscillations in the RestingConditionThere were no significant changes induced by EVS in both the mean and SDof the beta power the left motor, frontal and medial parietal regions duringrest (Figs. 5.7 and 5.8).5.4 DiscussionWe demonstrated EVS elicits a range of changes in task performance, withthe most remarkable results in the PDMOFF, followed by HC and PDMONgroups. The improvement was greater with EVS1 than EVS2 overall andcorrelated with more pronounced beta ERD in the left motor region duringmotor preparation and faster recovery of the beta power in the frontal andmedial parietal regions during motor execution. Finally, the modulatoryeffects on the beta oscillations were specific to performing the motor task.1005.4. DiscussionFigure 5.7: Effects of EVS1 on the beta power during the 60-s resting condition. The betapower in each of the 2-s epochs was computed using the multi-taper method (seven Slepiansequences; frequency resolution = 0.5 Hz), and then the temporal mean and SD acrossthe epochs in the EVS1 condition were compared to those in the sham condition. Bargraphs represents the % change in the temporal mean (left) and the SD (right) inducedby EVS1 for subjects in each group and error bars indicate SEM. Significant P valuesfrom one-sample t-tests are indicated (*P < 0.05 before multiple correction). To accountfor multiple comparisons, false discovery rate (FDR)-corrected P values were calculatedusing the method introduced by Benjamini and Hochberg [27] (mafdr.m in MATLAB).After the multiple correction, none of the P values reached significance level. (A) ThePDMOFF group. (B) The PDMON group. (C) The HC group5.4.1 Abnormalities in Motor Control in PDWe found greater peak grip pressure and squeeze velocity in the PDMOFFgroup. This may seem counterintuitive at first, because impairment of motorfunction in PD would make one believe the patients might have a weakergrip force. However, PD patients have comparable overall muscle strengthto control subjects [404]. In addition, de novo PD patients use abnormallylarge grip forces when they are lifting and static-holding an object [104],suggesting this may be an intrinsic feature of PD.In linear regression analyses (Table 5.3), another pathological character-istic of the grip force was found in the PDMOFF group. For the PDMONand HC groups, the grip force and squeeze velocity were negatively relatedwith reaction time, indicating they tend to produce more ballistic move-1015.4. DiscussionFigure 5.8: Effects of EVS2 on the beta power during the 60-s resting condition. The betapower in each of the 2-s epochs was computed using the multi-taper method (seven Slepiansequences; frequency resolution = 0.5 Hz), and then the temporal mean and SD acrossthe epochs in the EVS2 condition were compared to those in the sham condition. Bargraphs represents the % change in the temporal mean (left) and the SD (right) induced byEVS2 for subjects in each group and error bars indicate SEM. Significant P values fromone-sample t-tests are indicated (*P < 0.05 before multiple correction). After multiplecorrection, none of the p-values reached significance level. (A) The PDMOFF group. (B)The PDMON group. (C) The HC group.ments when the responses are faster. In contrast, this relationship was lostin the PDMOFF group. PD subjects tended to squeeze harder independentof reaction time, supporting the view that modulation of vigour is impairedin the PD subjects [129, 236].Table 5.3: Linear regression analysis to demonstrate a relationship between fast responsesand vigorous movements. P -values represent statistical significance of the slopes.PDMOFF PDMON HCy (reaction time, ms)x (peak grip pressure, a.u.)y = 473.0− 1.2x(P = 0.52)y = 468.3− 4.6x(P < 0.01)y = 518.4− 5.4x(P < 0.05)y (reaction time, ms)x (squeeze velocity, a.u.)y = 475.8− 0.2x(P = 0.24)y = 468.0− 0.5x(P < 0.01)y = 537.5− 1.0x(P < 0.01)1025.4. Discussion5.4.2 EVS effects on the behavioural indicesTo the best of our knowledge, there is only one prior study that reported EVSeffects on reaction time in a motor task [432]. In that study, the reactiontime of the PD patients was decreased by EVS without changes in omissionor commission errors, suggesting cognitive processes were not affected by thestimulation. Considering the task was rudimentary, the authors suggestedEVS may have improved bradykinetic motor execution. We found EVS im-proved the reaction time in both PDMOFF and HC groups, suggesting theunderlying neural mechanism may be common across the groups rather thanspecific to the disease. EVS effects were mild for the PDMON group com-pared to the other groups, suggesting interactions between the medicationand EVS. Possibly this was due to ceiling effects, as levodopa is alreadyknown to suppress pathological beta-band oscillations in PD [93, 126].We carefully evaluated whether or not there was an accumulated and/orlearning effect due to repetitive stimulation and task using repeated-measuresANOVA, and none of the behaviour measures showed a systematic differenceacross trials (Table 5.4). We conclude that our observations were consistentwith a stimulus-specific effect that was robust and repeatable as comparedto insignificant behaviour changes found using the other stimuli (not re-ported here). However, there were differences across sham trials, suggestingthat although the stimulation trials did not show any accumulated effect,there may have been some carryover during rest. More work is required todetermine the duration of any possible carryover effect at rest.5.4.3 Functional significance of beta ERD in voluntarymovementThe correlation between better performance and greater beta ERD is inline with the concept that augmented ERD reflects involvement of a largerneural network in information processing, which facilitates more efficienttask performance [291, 367]. The reduced development time and enhancedmagnitude of the ERD in the motor preparation period is similar to effectsof levodopa on PD subjects [90, 227], which may indicate facilitation ofreadiness of motor network for the upcoming movement [14, 15, 97, 127]. Inaccordance with this view, at the subcortical level, the early timing of betaERD onset in the STN is correlated with shorter reaction times [426].In the motor execution period, EVS was found to facilitate more rapidbeta power recovery in the frontal and medial parietal regions belongingto the frontoparietal network. The frontoparietal network is important in1035.4. DiscussionTable 5.4: Results of the repeated measures ANOVA to investigate accumulated and/orlearning effect on the behaviour measures.SS df MS F P valuePeak grippressureTimeGroup × TimeError (Time)5.97612.7250.5910204600.5980.6350.5451.0971.1660.3690.279Squeeze velocity TimeGroup × TimeError (Time)629.721716.838510102046062.9785.8483.720.7521.0250.6750.430Movement time TimeGroup × TimeError (Time)6799590161.279×1061020460679.92950.82780.40.2441.0610.9910.388Squeeze time TimeGroup × TimeError (Time)681.2641.524570102046068.1232.0853.411.2750.6000.2420.913Reaction time TimeGroup × TimeError (Time)8040.2194224016901020460804.02971.11873.230.9211.1120.5140.333Peak time TimeGroup × TimeError (Time)161652738541133010204601616.51369.3849.21.8081.5310.0570.066conscious motor intention and movement awareness [88], facilitating motorpreparation and execution through information flow between the parietaland frontal cortices [419]. In addition, the frontal region includes supplemen-tary motor area (SMA) and premotor areas (PMA) where beta oscillationsare known to play a critical role in motor control [290, 371]. The functionalsignificance of the rate of beta power recovery in these regions, however,still remains incompletely understood as most effort has been devoted tounderstanding the magnitude and spatial distribution. Nevertheless, wesuggest that it might be associated with post-movement resetting for thenext movement [365].1045.4. Discussion5.4.4 Modulation of beta ERD via NIBSSeveral tACS studies have demonstrated changes in the beta oscillations arecausally linked to motor behaviour (for review see [403]). Entraining betaoscillations in the motor cortex by 20-Hz tACS affected motor-evoked poten-tials (MEPs) [107] and slowed down voluntary movements [169, 298]. Ourresults extend these studies and show that modulating beta ERD appearscausally related to motor responses in the PD and HC subjects.A significant contribution of this work is to provide evidence of onlineeffects of EVS on beta ERD, which was not demonstrated in prior NIBSstudies presumably due to high-voltage artifacts from electrical stimulationseverely corrupting EEG. Here, we used stimulation frequencies beyond therange of cortical oscillations of interest in typical EEG studies (1–50 Hz) andmultisine signals that keep excitation power within specific frequency com-ponents [294], so stimulation artifacts in the EEG data could be effectivelyremoved by applying a digital filter.5.4.5 Beyond the modulation of cortical oscillations—potential mechanisms of EVSWe had two competing hypotheses for the EVS effects on the beta ERD.The first hypothesis was EVS reduces beta power overall, not necessarilyduring a motor task. This will facilitate movement as one must suppressthe beta power below a certain threshold before movement can commence[145]. The second hypothesis was EVS effects are dependent on behaviouralcontext so that vestibular inputs interact with movement-related signals inthe ventral thalamic region, the motor areas of the thalamus. The ven-tral thalamic region integrates multiple motor-related inputs and projectshighly refined motor plans back to the motor cortex. We did not find anysignificant changes in both the mean and SD of the beta power in duringrest (Figs. 5.7 and 5.8), supporting the latter hypothesis—namely involve-ment EVS effects were integrated with motor task commands rather thannonspecifically reducing overall beta oscillations.Previous studies support the notion of thalamic nuclei playing integra-tive and modulatory roles in sensorimotor processing. Vestibular nucleiin the brainstem have multiple ascending projections directly to the tha-lamus, primarily targeting the ventral anterior (VA), ventral lateral (VL),ventral posterior lateral (VPL), ventral posterior medial (VPM), intraminarnuclei and the geniculate bodies (Fig. 5.9A) [221, 247, 286], and strongactivations in these regions by vestibular stimulation have been reported1055.4. DiscussionFigure 5.9: Schematic representation of the major projections involved with motor func-tions. (A) Projections from the four vestibular nuclei to thalamic nuclei. The projectionsprimarily target at the ventral part of the thalamus, a region also known as the motor tha-lamus. The figure was adapted from [421] and modified based on [221, 368]. (SVN: superiorvestibular nucleus; MVN: medial vestibular nucleus; LVN: lateral vestibular nucleus; IVN:inferior vestibular nucleus; AN: anterior nucleus; LD: lateral dorsal nucleus; LP: lateralposterior nucleus; VA: ventral anterior nucleus; VL ventral lateral nucleus; VPL: ven-tral posterior lateral nucleus; VPM: ventral posterior medial nucleus; MD: mediodorsalnucleus; CM: centromedian nucleus; PF: parafascicular nucleus; LGN: lateral geniculatenucleus; MGN: medial geniculate nucleus) (B) Projections between the thalamus, BG andmotor cortices. The thalamus and layer V neurons in the motor cortices have reciprocalconnections [41]. The VA, VL and MD project to the putamen and caudate, and the VLand MD receive the bulk of BG outputs [136]. (Cd: caudate; Pu: putamen; GPe: globuspallidus externus; GPi: globus pallidus internus; STN: subthalamic nucleus; SNr: sub-stantia nigra pars reticulata; PMA: premotor cortex; SMA; supplementary motor cortex;M1: primary motor cortex) (C) An illustration of vestibular inputs influencing integrativeprocesses of a motor thalamic neuron. Inputs from the vestibular nuclei (VN) can betemporally and/or spatially integrated with thalamic afferents from other motor-relatedstructures such as the BG, modulating the neuronal activity.1065.4. Discussion[30, 55, 247, 366, 421], indicating a critical thalamic contribution to process-ing vestibular information [221, 421]. In particular, the ventral part of thethalamus has strong connections with motor-related structures such as M1,PMA, and BG (Fig. 5.9B) [41, 55, 286, 368] and its neural activities areassociated with a range of aspects in motor control [158, 194, 405], suggest-ing it serves as the motor thalamus. Recent studies highlight the functionalrole of the motor thalamus as a critical hub region to temporally and spe-cially integrate information required for controlling movement by efficientlyassigning weights to afferent inputs depending on the context and desiredmotor outcome and send this highly refined information to the motor cor-tices [41, 156, 421]. Thus, precise temporal and spatial pattern of the neuralactivation in the motor thalamus is critically involved with motor prepa-ration and execution, and we surmise that strong vestibular inputs to themotor thalamus had strong influence on the integrative process (Fig. 5.9C).This may also explain the mild EVS effects in the PD subjects on medicationas the inputs from the BG afferents to be integrated in the motor thalamusare a function of dopaminergic tone.We note that EVS might also affect the striatum, a region describedas an integrative centre for sensory information and involved in planningand execution of movements. Although the largest inputs to the striatumare from the cortex, recent studies have shed light on its subcortical path-ways critical to interpret and respond to environmental stimuli appropriately[109, 242]. Electrophysiological studies in animal models and neuroimagingstudies in humans have shown vestibular stimulation activates the head ofthe caudate nucleus and putamen [30, 42, 178, 234, 366], likely through theparafascicular thalamic nucleus [196, 368].Understanding the neural mechanisms behind EVS effects will providedeeper insights into brain functioning and brain-behaviour relationshipsand be crucial to improve neurotherapuetic effects. Although this studyreports electrophysiological evidence for its efficacy, findings from variousneuroimaging modalities together will allow for a more refined descriptionof EVS effects.107Chapter 6Spectral Clustering andDiscriminant CorrelationApproach to Estimation ofEVS Effects on FunctionalThalamic Subregions andBG–thalamic Connectivity inPD—fMRI StudyThe previous chapters have demonstrated therapeutic effects of EVS oncortical oscillations and motor behaviour in PD, which leads us to the nextquestion, “how does the modulation of vestibular nerve activity results inthe changes? What are the pathways and mechanisms underlying EVSeffects?”. Evidence from recent studies suggests that the thalamus maybe a key region involved in EVS effects as it has dense connections withboth the vestibular nuclei in the brain stem and basal ganglia and servesas a hub region to integrate and modulates sensorimotor information. Inthis chapter, we addressed the questions by investigating effects of EVS onthalamus activity and connectivity between the thalamus and basal gangliain PD.6.1 IntroductionPD is a neurological movement disorder characterized by several cardinalmotor symptoms (bradykinesia, rigidity, tremor, and postural instability)that are caused by substantial loss of dopamine in the substantia nigra parscompacta (SNc) located in the midbrain. Afferents from the SNc to the1086.1. Introductionbasal ganglia (BG) supply the striatum with dopamine [262], which is in-timately linked to mediating signals transmitted from the cerebral cortexto the thalamus, which, in turn, project back to the cerebral cortex (thecortical-BG-thalamic loops) via direct and indirect pathways [6, 199]. De-generation of the dopaminergic system results in striatal dopamine depletionof ∼44–98% in PD [313, 428], with changes found in various sub-regions ofthe striatum including not only the motor regions of the striatum (poste-rior and ventral putamen) receiving projections from the motor and premo-tor cortex but also the head of the caudate nucleus, the anterior putamenand the ventral striatum connected to the cerebral cortex, in particular thefrontal lobe [68, 271, 412].The BG work in conjunction with the thalamus and cortex to carry out anumber of segregated functions such as motor, oculomotor, cognitive, work-ing memory, and limbic processes in parallel [7, 135]. The primary role of thethalamus has been assumed to be the relaying of information, transferringthe signals from the output structures of BG (the internal segment of theglobus pallidus (GPi) and substantia nigra pars reticula (SNr)) to the cortex,but recent evidence suggests a much more active functional role of thalamiccell groups in processing BG information and subsequently modulation ofthe dynamics of cortical processing [135]. In PD, significant pathology inthe thalamus can be detected and contribute to parkinsonian motor dys-function [139]. In parkinsonian rodent models, administration of MPTPor 6-hydroxydopamine has been shown to induce degeneration of thalamicneurons projecting to the striatum as well as the loss of dopaminergic neu-rons in the SNc [19]. In addition, a range of intrinsic thalamic changes suchas bilateral morphological alterations [245], thalamic nuclei degeneration[46, 140, 149], and substantial deposition of α-synuclein [46, 149] have beenreported in PD patients. The thalamic changes consequently contribute toalterations in the dynamics of the cortico-BG-thalamic circuits [19] and canresult in aberrant connectivity patterns such as decreased connectivity ofthe thalamus with the GP, SN and the sensorimotor cortices [347].One promising way to modulate the cortico-BG-thalamic loop alteredin PD is through brain stimulation. Deep brain stimulation (DBS) of thethalamic ventral intermediate (Vim) nucleus, internal segment of the GPior the subthalamic nucleus (STN) has been shown to be effective in treat-ing parkinsonian tremor, dystonia and bradykinesia in PD [76, 310] withminimal effects on akinesia and gait disturbance [253]. One mechanismunderlying Vim DBS effects may be that the stimulation inhibits incom-ing afferent inputs carrying pathological signals or alters the excitability ofthalamic nuclei [12]. The success of DBS has led to strong interest in non-1096.2. Materials and Methodsinvasive brain stimulation (NIBS), which is particularly attractive due to itseasy accessibility and lower cost compared to DBS.EVS is a NIBS technique where electrical current is applied to mastoidprocess behind the ear to alter firing rates of vestibular afferents. Severalstudies have reported beneficial effects of random noise EVS on PD subjectsincluding improved motor performance [205, 278, 330, 432], enhanced pe-dunculopontine nucleus (PPN) connectivity [57], and modulation of corticaloscillations and connectivity strength [177, 204]. Such effects seem to beascribed to changes in neural dynamics in the brain by manipulation of as-cending pathways from the vestibular nuclei located in the brainstem to thecerebral cortex via thalamocortical vestibular system [221, 392, 421], whichhas been evidenced in simultaneous fMRI and alterating current EVS studies[30, 220, 366]. The vestibular nuclei have multiple projections to thalamicnuclei including the ventrobasal nuclei that receive strong inputs from theBG and project to primary and motor cortices [421] and the intralaminarnuclei that projects to the striatum [196, 368], suggesting that EVS maybe capable of modulating functional connectivity between the thalamus andBG structures.Here, we investigated the effects of noisy EVS (nEVS) and 1-Hz sine EVS(sEVS) on activity of thalamic subregions obtained from connectivity-basedparcellation and their connectivity with BG structures using resting-statefMRI (rsfMRI) data recorded from PD and age-matched control subjects.We demonstrate that the sizes of thalamic functional subregions are alteredin the PD subjects, and both nEVS and sEVS “normalize” them, i.e. makesthem closer to that of control subjects. We found the connectivity betweenthe left thalamus and left BG is attenuated in the PD subjects off-medicationcompared to controls and PD subjects on-medication. This alteration wasnormalized by EVS in a stimulus dependent manner.6.2 Materials and Methods6.2.1 SubjectsFifteen PD subjects (11 males, age: 65.7 ± 8.8 (mean ± SD) years) and15 age-matched healthy controls (12 males, age: 63.7± 9.6 years; HC) par-ticipated in the study (Table 6.1). All subjects did not have any reportedvestibular or auditory disorders. The PD subjects were classified as havingmild to moderate PD state (Hoehn and Yahr stage I–III) without atypicalParkinsonism or other neurological disorders. The PD subjects stopped tak-ing their normal levodopa medication at least 12 hours, and any dopamine1106.2. Materials and MethodsTable 6.1: Study cohort demographicsMeasure PD HCAge (years) 65.7± 8.8 63.7± 9.6Gender (n), male:female 11:4 12:3Disease duration (years) 3.5± 2.0 N/AHoehn and Yahr scale 1.6± 1.0 N/AUPDRS III 17± 13 N/ALevodopa Equivalent Daily Dose (mg) [383] 741.7± 564.2 N/AHandedness all right-handed all right-handedAffected side, right:left 9:6 N/AAffecte side was defined as summed left and right scores of UPDRS III 3.3-3.8 and 3.15-3.17. Subjects with equal scores for the left and right were not counted.agonists 18 hours prior to the experiment (off-medication; PDMOFF). Uni-fied Parkinsons Disease Rating Scale (UPDRS) Part III was accessed in off-medication condition. After the first fMRI scan in off-medication condition,they took their regular dose of levodopa (L-dopa) medication and rested foran hour before beginning the second scan (on-medication; PDMON). Therewas one scanning session for the HC subjects.All subjects were recruited from the Pacific Parkinsons Research Centre(PPRC) at the University of British Columbia (UBC) and provided written,informed consent prior to participation. The study was approved by theUBC Ethics Review Board.6.2.2 EVSA bipolar constant current DS5 stimulator (Digitimer Ltd., Hertfordshire,UK) was used to deliver an alternating current via two MR-compatible pre-gelled Ag/AgCl electrodes (Biopac Inc., Montreal, Canada) placed over themastoid process behind each ear. Digital signals of the EVS stimuli were firstgenerated on a PC with MATLAB (MathWorks, MA, USA) and were con-verted to analog signals via a NI USB-6221 BNC digital acquisition module1116.2. Materials and Methods(National Instruments, TX, USA), which subsequently passed to the stim-ulator in the console room with the output cable leading into the scanningroom through a waveguide. The twisted coaxial output cable included fourcustom-built inductance capacity filters spaced 20 cm apart and tuned forthe Larmor frequency (128 MHz).Two different stimuli were tested in the study. nEVS was pink noise withzero-mean and 1/f -type power spectrum between 0.1–10 Hz, and sEVS wasa 1 Hz sine wave. Since individuals have an inherently subjective perceptionof EVS, prior to scanning, we determined the individual sensory thresholdlevel (cutaneous sensation at the electrode site) utilizing systematic proce-dures used in prior EVS studies [205, 424] and delivered EVS at 90% of theindividual threshold level. Before conducting the real experiment, we carriedout a pilot study and recorded the current actually delivered to the subjects.We confirmed that the delivered current matched with the designed stimulusand device was compatible with MRI.6.2.3 MRI acquisition and preprocessingImaging data were acquired using a Philips Achieva 3.0T R3.2 scanner(Philips Medical Systems, Netherlands). Before scanning, all the subjectswere instructed to lie on their back in the scanner and had several minutesto acclimatize themselves to the scanner environment with the eyes closed.Before the functional scans, high-resolution T1-weighted images of the entirebrain were acquired (repetition time = 7.9 ms, echo time = 3.5 ms, flip angle= 8). For functional scans, BOLD contrast echo-planar (EPI) T2*-weightedimages (repetition time = 1985 ms, echo time = 37 ms, flip angle = 90, fieldof view = 240 mm × 240 mm, matrix size = 128 × 128, pixel size = 1.9 mm× 1.9 mm) were acquired after 4 initial dummy scans. The scan order waskept consistent for all subjects as sham, nEVS, and sEVS, and there was a2-min break between each condition to avoid any possible post-stimulationeffects. The sham condition was 8-min resting state and 5-min continuousstimulation was applied in the EVS conditions.6.2.4 Data preprocessingFunctional MRI data were preprocessed using DPABI version 3.0 [433] andSPM8 software package (https://www.fil.ion.ucl.ac.uk/spm). The first 5time points were discarded to allow the magnetization to approach a dy-namic equilibrium and to allow participants to get used to the scanningnoise. Then the images were corrected for slice timing effects, resliced to1126.2. Materials and Methods3.0 × 3.0 × 3.0 mm isotropic voxels, movement corrected using rigid bodyalignment, normalized into standard MNI space. The fMRI data were thenspatially smoothed with a 6-mm Gaussian kernel to increase its signal-to-noise ratio. To reduce the potential confounds of head motion and possi-ble effects of physiological artefacts, nuisance time courses were voxel-wiseregressed from the processed data to remove sources of variance includ-ing head-motion parameters, their temporal derivatives and their squares,white-matter signal and CSF signal. Any linear or quadratic trends wereremoved from fMRI signals. The fMRI data were finally bandpass filteredat 0.01 Hz to 0.08 Hz as recommended.6.2.5 Connectivity based parcellation of thalamusTo determine the regional specificity within the thalamus associated withthe disease and investigate their regional alterations by EVS, we performedconnectivity based subregional parcellation using Normalized Cut SpectralClustering (NCUT) algorithm that is robust to outliers [166], can easilyincorporate spatial constraints, and performs well on fMRI data [74].Suppose the number of voxels in the thalamus is N , and we construct agraph G = {V,E}, where the vertex set V represents all N voxels, and E isthe edge set. Let W denotes the weight matrix between vertices, and W (i, j)is defined as a function of correlation between nodes i and j. To divide thegraph into two disjoint sets A and B, we try to minimize the connectionsbetween two sets while maximizing the connections within each set, and theobjective function of NCUT is defined as,NC(A,B) =cut(A,B)assoc(A, V )+cut(A,B)assoc(B, V )(6.1)where cut(A,B) =∑i∈A,j∈BW (i, j) is the sum of weighted connectionsbetween sets A and B, and assoc(A, V ) =∑i∈A,j∈V W (i, j) is the totalweights of connections from nodes in A to all other nodes in the graph.The NCUT algorithm can be further extended to the K-way partitionwith K representing the number of partitions [38]. Let D be an N × Ndiagonal matrix with D(i, i) = di =∑Nj=1w(i, j), and indicator matrixY ∈ {0, 1}N ·K represents the partition of graph G. Then, if node i belongsto partition set j, Y (i, j) = 1, otherwise, Y (i, j) = 0. This optimizationproblem can be efficiently solved as a generalized eigenvalue problem,NC = K − Tr(Z ′(D− 12WD 12 )Z) (6.2)1136.2. Materials and Methodswhere Z ′Z = IK , and IK is the identity vector with length K. The solu-tion of Z is the matrix with the k eigenvectors associated with the first Keigenvalues of matrix D−12WD12 . Z can be considered as the new set ofcoordinates for the graph and we further apply the K-means to obtain thecluster indicator matrix Y .As we were interested in the spatially confluent parcellations, we incor-porated spatial information into the subregion parcellation of the thalamuswhere only spatially continuous voxels were allowed to be connected witheach other in the similarity matrix W . In addition, negative correlationswere removed from the network, resulting in the symmetric, positive andspatially continuous time-dependent similarity matrices for spectral cluster-ing. After the parcellation, the number and mean time course of the voxelsin each subregion were computed.6.2.6 Subregion size analysisMultivariate logistic regression analysis was used to compare sizes of the tha-lamic subregions between PDMOFF and HC groups (independent variables:the sizes of the bilateral thalamic subregions; dependent variable: group).The weights (wlogit) obtained from the logistic regression model were thenfurther utilized to compute thalamic subregion sizes of the PDMON sub-jects (sham) and PDMOFF subjects during stimulation to evaluate L-dopaand EVS effects.6.2.7 Thalamus-BG connectivity analysisFunctional connectivity between the thalamic subregions and BG structuresincluding the bilateral caudate, putamen and pallidum were computed usingPearson correlation and normalized between zero and one for each subject.We examined the ipsilateral and contralateral functional connectivity be-tween the thalamus and BG that is maximally different across the PDMOFF,PDMON, and HC by utilizing discriminant correlation analysis (DCA), afeature fusion method that incorporates discrimination of different classesinto a canonical correlation analysis (CCA)-based algorithm. The basicconcept of DCA is to search for transformation matrices (WX and WY ) toproject original data sets (X and Y ) into a space where the new projecteddata (X ′ and Y ′) are correlated with each other while simultaneous classseparation is achieved (for detailed description of the algorithm, see [137]).Here, we created one data set, X (n×p matrix; n = number of subjects; p =1146.2. Materials and Methodsnumber of ipsilateral connectivity), by concatenating the functional connec-tivity between the left BG structures and left thalamic subregions across allsubjects in the sham condition. Another data set, Y (n×q matrix; q = num-ber of contralateral connectivity), was created by concatenating functionalconnectivity between the left BG structures and right thalamic subregionsacross all subjects in the sham condition. The DCA model is therefore,X ′ = XWXY ′ = YWY(6.3)where WX (p × k) and WY (q × k) are the weight vectors, and X ′ (n × k)and Y ′ (n×k) are the transformed ipsilateral and contralateral connectivity,respectively. We used the first columns of X ′ and Y ′ that have the maximumcorrelation and the corresponding weights in further analyses.The obtained weights and transformed connectivity were used to exam-ine discriminant connectivity patterns and differences in the connectivitystrengths between the PD and HC groups at baseline. Likewise, DCA wasapplied to the functional connectivity matrices between the right BG struc-tures and subregions in the right and left thalamus, and the weights andtransformed ipsilateral and contralateral connectivity were further analyzed.Finally, the computed weights (WX and WY ) were used in the subsequentanalyses to infer EVS effects. That is, functional connectivity matrices werecreated in the same manner as the above using the fMRI data acquired inthe EVS conditions and multiplied by the transformation weights obtainedfrom the sham condition. The transformed connectivity (X ′ and Y ′) duringeach EVS were then compared with those obtained from the sham data.6.2.8 Statistical analysisDifferences between groups in the thalamus functional subregion size andthe BG-thalamic connectivity from the DCA were assessed using one-wayANOVA. EVS effects on the subregion size and the BG-thalamic connec-tivity within a group were evaluated using repeated measures ANOVA withstimulation condition (sham, nEVS and sEVS) as the within-subject fac-tor. Statistical significance was considered when P values were < 0.05 afterBonferroni correction for multiple comparison.1156.3. Results6.3 Results6.3.1 Thalamus ParcellationEach thalamus was segmented into five subregions by applying the spatiallyconstrained normalized cut approach (Fig. 6.1). According to the vari-ance ratio criterion, the data-driven averaged optimal number of clusterswas three. However, anatomically, the thalamus is classically segmentedinto a number of relay, association, and nonspecific nuclei, including themedial dorsal nucleus, an anterior nuclear group, the ventral nuclear group(ventral anterior, ventral lateral, ventral posterolateral and ventral postero-medial nuclei), the lateral nuclear group and the pulvinar nucleus, howeverat the relatively coarse spatial resolution of fMRI, these may not be indi-vidually discriminable. In a previous rsfMRI study, five distinct regions,including ventral anterior nuclei, ventral lateral nuclei, pulvinar, anteriornuclei and medial dorsal nuclei were identified based on independent com-ponent analysis [176]. Using diffusion tensor parcellation approach, Kumarand colleagues [192] also selected five stable subunits in the anterior, medial,lateral-anterior, lateral-posterior and posterior thalamus. Six subunits havealso been chosen to investigate their subregional connectivity [35, 164]. Here,we chose the number of subregions in thalamus to be five as an optimal bal-ance between ease of interpretation of the results and complexity of the sub-regional structures consistent with anatomical prior knowledge and previousstudies. The identified functional subregions we segmented were related tothe following anatomical thalamic nuclei as follows (Fig. 6.1): pulvinar andlateral posterior nuclei (PU+LP; subregion 1), anterior nuclear group andlateral dorsal nuclei (AN+LD; subregion 2), ventral postero-lateral nuclei(VPL; subregion 3), medial nuclei (MN; subregion 4), and ventral anteriorand ventral lateral nuclei (VA+VL; subregion 5).6.3.2 Thalamic subregion sizesA significant difference in the thalamus functional subregion size was foundacross the groups (F(2, 42) = 12.37, P < 0.001; Fig. 6.2A). The subre-gion size was significantly different between the PDMOFF and HC groups(P < 0.001), which was normalized by L-dopa medication (P < 0.001).The weights from the logistic regression analysis (wlogit) indicated that thePDMOFF group particularly had decreased sizes of the PU+LP in the leftthalamus and MN in the right thalamus (L1 and R4 in Fig. 6.2B) comparedto the other groups.1166.3. ResultsFigure 6.1: An example of thalamus parcellation results from a subject is displayed onthe horizontal slices arranged from superior to inferior (from left to right). The identifiedsubregions are colour coded and labeled with numbers (1: pulvinar and lateral posteriornuclei (PU+LP); anterior nuclear group and lateral dorsal nuclei (AN+LD); 3: ventralpostero-lateral nuclei (VPL); 4: medial nuclei (MN); 5: ventral anterior and ventral lateralnuclei (VA+VL))The stimulation condition showed significant effects on the subregionsize in the PDMOFF subjects (F(2, 28) = 3.85, P < 0.05; Fig. 6.2C). Bothstimuli normalized the thalamic subregion size in the PDMOFF subjects(P < 0.05) and the effect of nEVS was greater compared to sEVS. In con-trast, the stimulation did not show significant effects on the subregion sizefor PDMON (F(2, 28) = 2.03, P = 0.15) and HC (F(2, 28) = 3.12, P = 0.06)groups.6.3.3 Connectivity between the left BG and thalamicsubregionsThe ipsilateral connectivity between the left BG and left thalamus showeda significant difference across the groups (F(2, 42) = 3.6, P < 0.05; Fig.6.3A). The PDMOFF subjects demonstrated decreased ipsilateral connec-tivity compared to the HC group (P < 0.01), and L-dopa medication did notinduce a significant normalizing effect. The DCA weights indicated the fol-lowing ipsilateral connectivity is primarily reduced in the PDMOFF group:AN+LD region and putamen, AN+LD region and pallidum, VA+VL regionand pallidum, VA+VL region and putamen, and PU+LP region and puta-men. In contrast, the contralateral connectivity between the left BG andright thalamus did not show a significant difference across the group (F(2,42) = 1.87, P = 0.17).EVS significantly modulated the ipsilateral connectivity in the PDMOFF(F(2, 28) = 3.76, P < 0.05) and HC (F(2, 28) = 4.65, P < 0.05) groups (Fig.6.3B) whereas no significant EVS effect was found in the PDMON group(F(2, 28) = 1.23, P = 0.31). nEVS increased the ipsilateral connectivity inthe PDMOFF group (P < 0.05) while both nEVS and sEVS reduced theipsilateral connectivity in controls (P < 0.05 and P < 0.01, respectively).1176.3. ResultsFigure 6.2: Comparison of the transformed subregional size obtained from the logistic re-gression and EVS effects. (A) Comparison of the baseline (i.e., sham condition) subregionsizes between the PDMOFF, PDMON and HC. (B) The weights (wlogit) from the logisticregression analysis (L/R: left/right thalamus; 1–5: subregion index shown in Fig. 6.1).(C) Normalizing effects of EVS on the thalamic subregion size in the PDMOFF group.Significant P values are indicated (*P < 0.05; ***P < 0.001).The difference between the ipsilateral and contralateral connectivity strengthwas computed for each subject to examine symmetry of the left BG and tha-lamus interactions. We found a significant difference in the baseline symme-try across the groups (F(2, 42) = 4.24, P < 0.05; Fig. 6.3C). The PDMOFFgroup had a significantly weaker ipsilateral connectivity than the contralat-eral connectivity compared to the HC group (P < 0.01). This asymmetrywas normalized by EVS (P < 0.001 and P < 0.01 for nEVS and sEVS, re-spectively; Fig. 6.3D). For the 15 PD subjects, we found improvement in theasymmetry for 12 subjects by L-dopa medication (binomial test, P < 0.01),13 subjects by nEVS (binomial test, P < 0.001), and 12 subjects by sEVS(binomial test, P < 0.01) (Fig. 6.3E).1186.3. ResultsFigure 6.3: DCA results for the connectivity between left BG and bilateral thalami andEVS effects. (A) The weights and transformed data for the ipsilateral (left BG-left tha-lamus; top) and contralateral (left BG-right thalamus; bottom) connectivity in the shamcondition. (B) EVS effects on the ipsilateral connectivity. (C) Ipsilateral and contralateralconnectivity difference in the sham condition. (D) Effects of EVS on the ipsilateral andcontralateral connectivity difference. (E) Effects of L-dopa medication and EVS on theconnectivity difference for the 15 PDMOFF subjects. Significant P values are indicated(*P < 0.05; **P < 0.01; ***P < 0.001)1196.4. Discussion6.3.4 Connectivity between the right BG and thalamussubregionsThere was no significant difference in the ipsilateral connectivity (right BGand right thalamus F(2, 42) = 1.22, P = 0.31) and the contralateral con-nectivity (right BG and left thalamus; F(2, 42) = 1.62, P = 0.21) betweenPDMOFF, PDMON and HC groups in the sham condition. The stimulationdid not induce significant effects on the ipsilateral connectivity for all threegroups (PDMOFF: F(2, 28) = 0.71, P = 0.50; PDMON: F(2, 28) = 0.05,P = 0.96; HC: F(2, 28) = 3.10, P = 0.06). The contralateral connectiv-ity for PDMOFF and PDMON groups did not change by the stimulation(PDMOFF: F(2, 28) = 1.12, P = 0.34; PDMON: F(2, 28) = 0.70, P = 0.50)whereas both nEVS and sEVS decreased the contralateral connectivity inHC group (nEVS: P < 0.05; sEVS: P < 0.01).6.4 DiscussionThe main results of this study are threefold. We demonstrate that: 1)functional subregions of the thalami are subdivided differently between PDgroups on and off dopaminergic medication and healthy controls; 2) a signif-icant asymmetry exists in thalamic/BG interactions in the left BG, and, 3)these alterations are partially ameliorated with EVS in a stimulus-dependentmanner. Thus, this work provides additional mechanisms through whichEVS may prove beneficial in PD.6.4.1 Functional thalamic subregion sizes altered in PDStructurally, the overall volume of the thalamus as a whole appears to berelatively spared in PD, while microscopically, there can be selective degen-eration and structural changes of particular thalamic nuclei [122, 245]. Wedemonstrated that the PDMOFF group had a decreased size of the func-tional PU+LP subregion than the other groups. The PU plays an importantrole in visual perception, visual attention and visual target selection [98] andis critically involved in maintaining and modulating dynamics of neuronaloscillations in the visual cortex [197], which may be implicated in visualdysfunctions in PD [349, 373, 416]. Impaired visual functions in PD are re-flected by deficits in sensitivity to visual stimuli such as colour-contrast andluminance and reduced attentive visual processing [201, 216, 304]. Delaysin visually evoked potentials in PD have been also reported in electrophysi-ological studies [143]. The LP region is important in sensory perception as1206.4. Discussionit integrates sensory information and projects to superior parietal region inconcert with the PU. Recently, a voxel-based morphometry study has shownthat PD patients have reduced functional connectivity between the parietalregion and thalamus involved in visual and sensorimotor networks comparedto healthy controls [133], which may be associated with the reduced size ofPU+LP found in our results.We demonstrated that PDMOFF subjects also had a decreased func-tional size of the MN region. The MN region is known to have substan-tial reciprocal interconnections with the prefrontal cortex (PFC), playing acritical role in memory and various cognitive tasks [251]. Causal relation-ships between the MN and PFC have been demonstrated in lesion studiesin human and animal models (for reviews, see [280]) and recently it wasdemonstrated that decreased MN activity disrupts modulation of MN-PFCsynchrony required for working memory [281]. Although the MN plays akey role in working memory and behaviour flexibility that are recognized ascommon non-motor complications of PD [415], there is a lack of prior stud-ies that studied the functional role of the MN in PD. There are currentlyonly two studies available that reported significant white matter changes ofthe MN in PD patients, in relation to depression [59, 215]. Further studiesare required to elucidate implications of the volumetric changes of the MNregion in PD for cognitive functions.6.4.2 Asymmetric connectivity of left BG and thalamusAnatomical connectivity between the BG and bilateral thalami has beendemonstrated in animal models of PD where contralateral projections of theGP (primarily targeting the VM) and SNr (primarily targeting intralami-nar nuclei) were found [61, 157]. Bilateral GABAergic and glutamate path-ways have been reported [239], and a recent optogenetic study demonstratedthat stimulation of D1 and D2 dopamine receptors in the striatum activatesbilateral thalami [203]. Our results showed reduced connectivity betweenipsilateral left BG and thalamus connectivity in the PD group, which is inaccordance with prior studies demonstrating the SN and ipsilateral thalamusconnectivity is decreased in PD [248, 311, 347].We found significantly greater asymmetry in the connectivity betweenthe left BG and bilateral thalamus in the PDMOFF group, which becamemore symmetric with dopaminergic medication. This asymmetry was notobserved with the right BG, suggesting the left BG connectivity with thethalamus is more susceptible to disease effects. Prior studies have impliedthat left nigrostriatal pathway is more affected by the disease than the1216.4. Discussionright [303, 336], increasing the susceptibility of the left nigrostriatal network[53, 229]. This may be due to the handedness [22, 32, 69, 235, 338, 391, 399]although there may be other confounding factors. In our subjects, all PDsubjects were right-handed, whereas the obvious clinical asymmetry (de-fined as the difference between the summed UPDRS scores of the left andright extremities with respect to rigidity, bradykinesia and tremor [39]) wasnot found (left worse: 40%; Table 6.1). While our results are consistentwith the notion that the asymmetric left BG-thalamus connectivity is dueto handedness, we do not have sufficient evidence to prove/disprove thishypothesis.6.4.3 Potential mechanism of EVSWe demonstrated that nEVS and sEVS both significantly normalized func-tional subregion size in the PDMOFF group. Additionally, nEVS enhancedthe diminished ipsilateral connectivity between the left BG and thalamusin the PDMOFF group. The 1/f type power spectrum of nEVS reflectsthe power distribution found in cortical and subcortical functional networks[56], and its beneficial effects in PD have been demonstrated in previousstudies [205, 278, 424, 432]. The stochastic resonance phenomena (alsoknown as “stochastic facilitation” in biological and medical fields) wherea sub-threshold random stimulus elicits functional benefits in a non-linearsystem [238] such as the nervous system has been proposed as a mechanismto explain how the randomly-varying stimuli may provide beneficial effects.Anatomically, the vestibular nuclei in the brainstem have multiple pro-jections directly to the thalamus including the VA, VL, VPL, VPM, in-tralaminar nuclei, and geniculate bodies [221, 247, 286]. Prior fMRI-EVSstudies have shown that vestibular stimulation induces strong activation inthe thalamus [30, 55, 247, 366, 421]. However, neither effects on the thalamicsubregions nor modulatory influences on the BG-thalamic functional con-nectivity have been previously described. Our findings provide evidence thatEVS modulates subgroups of thalamic nuclei and interaction with the BGin a stimulus-dependent manner. We elucidated the EVS effects focusing onthe thalamus and BG based on their anatomical importance in both PD andvestibular information processing. Further studies to investigate effects atthe cortical level with respect to the changes at the subcortical level shownhere will provide deeper understanding in the neurotherapeutic mechanismof EVS. The results from the current study suggests that “normalization”of disrupted BG-thalamus connectivity may be a key mechanism throughwhich EVS induces beneficial effects in PD.122Chapter 7Conclusion and Future WorkIn this dissertation, in an effort to advance application of EVS as a potentialtherapeutic intervention for PD, we utlized new multisine EVS stimuli andinvestigated effects of the different stimuli on brain activity and motor func-tion in PD and healthy subjects. In addition, we investigated effects of noisyEVS, the stimulus type used in the majority of prior EVS studies on PD,on motor function in PD subjects. By utilizing a new motor task and ana-lytical methods, we have added valuable new information on top of existingfindings. Furthermore, we conducted simultaneous EVS-fMRI experimentsto probe the fundamental mechanisms of EVS utilizing the excellent spatialresolution of the fMRI data. This is the first fMRI study to investigateeffects of EVS in PD, and we believe the outcomes from the study will sig-nificantly increase our understanding of the EVS mechanisms. Finally, wedeveloped a novel denoising method to remove stimulation artifacts in EEG,which has been identified a critical challenge to resolve in order to be ableto investigate immediate stimulation effects on brain oscillations.7.1 Conclusion and Summary7.1.1 Conclusion and SummaryIn Chapter 2, subthreshold noisy EVS was applied to PD subjects whilethey performed a visuomotor joystick tracking task, which alternated be-tween 2 task conditions depending on whether the displayed cursor positionunderestimated the actual error by 30% (‘Better’) or overestimated by 200%(‘Worse’). Coefficients from LDA indicated that noisy EVS had significantlyaffected the perceived error between the target and the displayed cursor po-sition, displayed cursor velocity and acceleration. It was also found thatEVS made the subject more smoothly track the target as observed in SNRand less overshoot in the tacking. These results in accordance with previousfindings that noisy EVS has effects on modulating motor functions in PD,and raises the additional question as to whether the effects are specific onlyto the type of stimulus used in this study (0.1–10 Hz, pink noise) or can1237.1. Conclusion and Summarybe induced, through modulation of the sensorimotor system via stochasticfacilitation or other neural mechanisms, with other types of stimuli (e.g.,sine waves, white noise).In Chapter 3, a new framework was proposed for removal of stimula-tion artifacts in EEG recordings using a JBSS technique. As opposed toconventional methods (PCA and ICA), which decompose a single datasetinto individual components, this new approach is able to simultaneously ac-commodate multiple datasets and identify source components using theircorrelation or independence within and between datasets. It was demon-strated that the proposed method, q-IVA, outperforms the conventionalmethods, MCCA and IVA in simulations. When examining real data, q-IVA successfully attenuated the stimulation artifact and enabled detectionof physiological changes in the cleaned EEG data. The results suggest thatq-IVA is an effective denoising method for the investigation of neurophysio-logical online effects of EVS and the method is utilized in the simultaneousEVS-EEG study in Chapter 4.For the evaluation of multisine EVS stimuli on brain oscillations, inChapter 4 discriminant features in widespread cortico-cortical couplings be-tween PD and HC groups were first identified using SDA and the changesin the direction and magnitude of the discriminant features induced by theEVS stimuli were examined. It was demonstrated that the discriminantfeatures are associated with the strength of cortical couplings in the sen-sorimotor region, and variability and complexity of the coupling dynamicsin predominantly theta and alpha bands. The discriminant features in thePD subjects were found to modulated by 4-8 Hz, 50-100 Hz and 100-150 HzEVS such that during and after each stimulation the features were broughtclose to those of the HC subjects. The direction of the changes (normaliz-ing or worsening) induced by the stimulation was same across the differentstimuli while the magnitude and duration of the aftereffects were stimulus-dependent.In Chapter 5, effects of multisine EVS stimuli on motor functions wereexamined using joint analysis (DCA) of the EEG and behaviour data thatwere recorded simultaneously from PD and HC subjects while they were per-forming a squeeze-bulb motor task. First, it was demonstrated that both50-100 Hz and 100-150 Hz multisine EVS improved task performance of thePD subjects when they were off-medication and induced less improvementwhen they were on medication, suggesting interaction between the medica-tion and EVS. The results derived from the DCA demonstrated that theimprovement in the task performance was correlated with the EVS-inducedchanges in the magnitude and dynamics of beta ERD in the left motor,1247.1. Conclusion and Summarybroad frontal and medial parietal regions. The effects of EVS were foundgreater with the 100-150 Hz stimulation in the PD subjects off medicationand HC subjects as compared to the 50-100 Hz stimulation and the PDsubjects already on optimal medication.The focus of Chapter 6 was to provide a deeper understanding of thefundamental mechanisms underlying the EVS effects shown in PD, focusingon the thalamus based on its anatomical importance in both motor networksand vestibular information processing. FMRI data were collected from thePD and HC subjects while 0.1-10 Hz noisy EVS (same as in Chapter 2)and 1-Hz sinusoidal EVS were continuously being applied for 5 minutes.The results demonstrate that both EVS significantly normalized the sizeof PU+LP functional subregions in the PDMOFF group and the size ofVPL subregion in the PDMON group. In addition, the noisy EVS waseffective in improving the connectivity strength between the left BG andbilateral thalami such that the strength of the ipsilateral and contralateralconnectivity, was less asymmetric, as would be seen in HC subjects. Thefindings suggest that modulation of thalamo-BG connectivity may be onepotential mechanism underlying EVS effects on motor improvement in PD.In conclusion, this dissertation aims to improve our understanding ofEVS technique as a potential therapeutic intervention for PD. The resultsdemonstrated that EVS is effective in modulating brain oscillations, activ-ity and functional connectivity of the thalamus and improving motor be-haviours in PD, and by varying stimulation waveforms the effect size andduration can be further improved. The presented work lays the groundworkand demonstrates a potential to further develop EVS for a patient-specificneuromodulation tool to improve motor functions in PD.The contributions of this dissertation are summarized as follows:1. Based on approaches used to describe behaviour of the system in thefield of system identification, novel multisine stimuli for EVS weredeveloped such that they have advantages over conventional noisy EVSin that they can excite specific components of responses, minimizeunwanted loss of accuracy (leakage) in the frequency domain, applymaximum power to the system and improve subjects discomfort.2. Multisine EVS signals were tested for the first time with neuroimagingmodality tol their investigate effects on electrical brain activities.3. It was first demonstrated that EVS can modulate phase-based corticalcouplings. Results indicated that EVS improve both strength andtemporal dynamics of cortical couplings in PD.1257.1. Conclusion and Summary4. Results demonstrated that EVS significantly improves motor task per-formance of PD subjects off-medication.5. The correlation between changes in brain oscillations and motor be-haviours induced by EVS was demonstrated for the first time. Re-sults indicated that EVS can modulate the magnitude and dynamicsof movement-related desynchronization of beta oscillations, which arecorrelated with improvement of the motor task performance.6. Comparison of EVS effects on cortical couplings, movement-relatedbeta desynchronization, and motor task performance between off- andon-medication conditions of the PD subjects indicated interactionsbetween L-dopa medication and EVS. It was found that, in general,improvements by multisine EVS is greater in the off-medication con-dition compared to the on-medication condition.7. It was found that EVS effects can be dependent on brain states. Re-sults demonstrated that EVS modulated the power of beta oscillationswhen the subjects were engaged in the motor task but did not changethe beta power when subjects were at rest.8. EVS was shown to change activities of the functional thalamic subre-gions and the connectivity between the thalamus and BG in PD. Thisprovides a potential mechanism of how EVS can affect motor systemsin PD.9. A novel method was devised for removal of stimulation artifacts inEEG. The results demonstrated a robust artifact removal performanceof q-IVA method through the simulation and real-data studies. Thus,the method could be a promising tool to properly analyze the EEGacquired during electrical brain stimulation in future studies.7.1.2 Limitations and suggestions for future workDevelopment, implementation and evaluation of EVS is highly multidisci-plinary work, and accordingly a number of interesting areas of research canbe suggested to further improve the current EVS technique as follows:1. Comparison of the noisy EVS and multisine EVS: In Chapters4 and 5, the results demonstrated promising efficacy of multisine EVSto improve motor functions in PD. As most prior EVS studies in PDhave used a noisy stimulus, it would be informative to know how mul-tisine EVS compares to a noisy stimulus in the investigated results or1267.1. Conclusion and Summaryin future studies. Synchronous neuronal activities between brain re-gions at specific frequencies are involved to carry out particular brainfunctions, and it has been shown that changes in brain activities andresultant outcomes are dependent on the frequency of the stimulationapplied. Therefore, multisine EVS bounded in a specific frequencyband would be likely to bring about changes in particular brain re-gions or network compared to a noisy stimulus. What would happenif the frequency band of a multisine signal is a subset of the one ofthe noisy signal like as in this study? (e.g., the 4–8 Hz multisine and0.1–10 Hz pink noise). Would be the effect size greater with the mul-tisine compared to the pink noise? To be able to answer this questionin the future would be critical to for providing rationales for buildingoptimized stimulation signals that can enhance the effect size of thestimulation.2. Further validation of q-IVA method: Q-IVA method was testedin the simulation study where the stimulation artifacts were derivedfrom the multisine signal in 4-8 Hz. In practice, stimulation signals canbe in various waveforms (e.g., sine waves, chirps, multisines, pulses)and it is recommended to test out the method with different types ofstimulation artifacts and identify when q-IVA works the best and whenit is less successful. This information will help promote the NIBS fieldby providing more freedom for selecting stimulation parameters thathave the available denoising option to probe online stimulation effects.3. Investigation of relationship between brain states and re-sponses to EVS: Although this research has focused on the EVSeffects at group levels (e.g., PD patients vs. healthy controls), it is ofgreat interest to determine how and why the effects vary across differ-ent individuals. This has partially done in this thesis by looking at thecorrelation between effect sizes of EVS and clinical characteristics ofthe PD subjects such as disease duration, severity and the amount ofdopaminergic medication taken. Nevertheless, relationships betweencharacteristics of individual brain activities at the baseline and EVSeffect sizes in both PD patients and healthy controls still remain elu-sive. Brain activities can be characterized in many different ways,ranging from conventional power spectral density of a single channel tothe complex non-stationary dynamics of multiple channels simultane-ously. High-dimensional feature spaces will likely need to be searchedfor to find the individually–specific EVS effect. State-of-the-art ma-1277.1. Conclusion and Summarychine learning approaches are warranted to address this subject.4. Investigation of nonlinear effects: The results from Chapters 4and 5 indicate presence of nonlinear effects of EVS in that high-frequency stimulation (50–150 Hz) can significantly influence low-frequency(1–50 Hz) cortical oscillations, which has not been previously reportedelsewhere. Moreover, it was shown in Chapter 5 that the non-lineareffects are also dependent on the context (i.e., resting vs. movement),suggesting that the underlying vestibular information processing isinfluenced by several integrative processes in the brain. Future in-vestigation with more finely tuned stimulation frequency parameterscombined with EEG and/or fMRI is warranted to elucidate the non-linear effects of EVS.128Bibliography[1] D. Aarsland, K. Bronnick, C. Williams-Gray, D. Weintraub,K. Marder, J. Kulisevsky, D. Burn, P. Barone, J. Pagonabarraga,L. Allcock, G. Santangelo, T. Foltynie, C. Janvin, J. P. Larsen, R. A.Barker, and M. Emre. Mild cognitive impairment in Parkinson dis-ease: A multicenter pooled analysis. Neurology, 75(12):1062–1069, 92010.[2] P. P. Acharjee, R. Phlypo, L. Wu, V. D. Calhoun, and T. Adali. In-dependent Vector Analysis for Gradient Artifact Removal in Concur-rent EEG-fMRI Data. IEEE transactions on bio-medical engineering,62(7):1750–8, 7 2015.[3] M. Adib and E. Cretu. Wavelet-based artifact identification and sep-aration technique for EEG signals during galvanic vestibular stimu-lation. Computational and Mathematical Methods in Medicine, 2013,2013.[4] S. Ahn, S. E. Zauber, R. M. Worth, T. Witt, and L. L. Rubchinsky.Interaction of synchronized dynamics in cortex and basal ganglia inParkinson’s disease. European Journal of Neuroscience, 42(5):2164–2171, 9 2015.[5] A. D. Akkaya and M. L. Tiku. Robust estimation in multiple linearregression model with non-Gaussian noise. Automatica, 44(2):407–417,2 2008.[6] G. E. Alexander and M. D. Crutcher. Functional architecture of basalganglia circuits: neural substrates of parallel processing. Trends inneurosciences, 13(7):266–71, 7 1990.[7] G. E. Alexander, M. R. DeLong, and P. L. Strick. Parallel Organi-zation of Functionally Segregated Circuits Linking Basal Ganglia andCortex. Annual Review of Neuroscience, 9(1):357–381, 3 1986.129Bibliography[8] P. J. Allen, O. Josephs, and R. Turner. A method for removing imag-ing artifact from continuous EEG recorded during functional MRI.NeuroImage, 12(2):230–9, 8 2000.[9] F. Alonso-Frech. Slow oscillatory activity and levodopa-induced dysk-inesias in Parkinson’s disease. Brain, 129(7):1748–1757, 7 2006.[10] J. L. Amengual, M. Vernet, C. Adam, and A. Valero-Cabre´. Localentrainment of oscillatory activity induced by direct brain stimulationin humans. Scientific Reports, 7(1):41908, 12 2017.[11] M. Anderson, X.-L. Li, and T. Adal. Nonorthogonal IndependentVector Analysis Using Multivariate Gaussian Model. pages 354–361.Springer, Berlin, Heidelberg, 2010.[12] T. R. Anderson, B. Hu, K. Iremonger, and Z. H. T. Kiss. Selec-tive Attenuation of Afferent Synaptic Transmission as a Mechanismof Thalamic Deep Brain Stimulation-Induced Tremor Arrest. Journalof Neuroscience, 26(3):841–850, 1 2006.[13] F. G. Andres and C. Gerloff. Coherence of sequential movements andmotor learning. Journal of clinical neurophysiology : official publica-tion of the American Electroencephalographic Society, 16(6):520–7, 111999.[14] A. G. Androulidakis, L. M. F. Doyle, T. P. Gilbertson, and P. Brown.Corrective movements in response to displacements in visual feedbackare more effective during periods of 13-35 Hz oscillatory synchrony inthe human corticospinal system. European Journal of Neuroscience,24(11):3299–3304, 12 2006.[15] A. G. Androulidakis, L. M. F. Doyle, K. Yarrow, V. Litvak, T. P.Gilbertson, and P. Brown. Anticipatory changes in beta synchrony inthe human corticospinal system and associated improvements in taskperformance. European Journal of Neuroscience, 25(12):3758–3765, 62007.[16] D. E. Angelaki and K. E. Cullen. Vestibular System: The Many Facetsof a Multimodal Sense. Annual Review of Neuroscience, 31(1):125–150,7 2008.[17] A. Antal, K. Boros, C. Poreisz, L. Chaieb, D. Terney, and W. Paulus.Comparatively weak after-effects of transcranial alternating current130Bibliographystimulation (tACS) on cortical excitability in humans. Brain Stimu-lation, 1(2):97–105, 4 2008.[18] A. Antal and W. Paulus. Transcranial alternating current stimulation(tACS). Frontiers in human neuroscience, 7:317, 2013.[19] M. S. Aymerich, P. Barroso-Chinea, M. Pe´rez-Manso, A. M. Mun˜oz-Patin˜o, M. Moreno-Igoa, T. Gonza´lez-Herna´ndez, and J. L. Lanciego.Consequences of unilateral nigrostriatal denervation on the thalamos-triatal pathway in rats. European Journal of Neuroscience, 23(8):2099–2108, 4 2006.[20] B. Ballanger, S. Thobois, P. Baraduc, R. S. Turner, E. Broussolle,and M. Desmurget. Paradoxical Kinesis is not a Hallmark of Parkin-son’s disease but a general property of the motor system. MovementDisorders, 21(9):1490–1495, 9 2006.[21] A. T. Barker. An introduction to the basic principles of magnetic nervestimulation. Journal of clinical neurophysiology : official publicationof the American Electroencephalographic Society, 8(1):26–37, 1 1991.[22] M. J. Barrett, S. A. Wylie, M. B. Harrison, and G. F. Wooten. Hand-edness and motor symptom asymmetry in Parkinson’s disease. Journalof Neurology, Neurosurgery & Psychiatry, 82(10):1122–1124, 10 2011.[23] R. J. Barry, A. R. Clarke, S. J. Johnstone, C. A. Magee, and J. A.Rushby. EEG differences between eyes-closed and eyes-open restingconditions. Clinical Neurophysiology, 118(12):2765–2773, 12 2007.[24] G. Batsikadze, V. Moliadze, W. Paulus, M.-F. Kuo, and M. A. Nitsche.Partially non-linear stimulation intensity-dependent effects of directcurrent stimulation on motor cortex excitability in humans. The Jour-nal of Physiology, 591(7):1987–2000, 4 2013.[25] B. S. Baxter, B. Edelman, X. Xiaotong Zhang, A. Roy, and B. Bin He.Simultaneous high-definition transcranial direct current stimulation ofthe motor cortex and motor imagery. In 2014 36th Annual Interna-tional Conference of the IEEE Engineering in Medicine and BiologySociety, volume 2014, pages 454–456. IEEE, 8 2014.[26] A. Belouchrani, K. Abed-Meraim, J.-F. Cardoso, and E. Moulines. Ablind source separation technique using second-order statistics. IEEETransactions on Signal Processing, 45(2):434–444, 1997.131Bibliography[27] Y. Benjamini and Y. Hochberg. Controlling the false discovery rate:a practical and powerful approach to multiple testing. Journal of theRoyal Statistical Society, 57(1):289–300, 1995.[28] D. H. Benninger and M. Hallett. Non-invasive brain stimulation forParkinsons disease: Current concepts and outlook 2015. NeuroReha-bilitation, 37(1):11–24, 8 2015.[29] D. H. Benninger, M. Lomarev, G. Lopez, E. M. Wassermann, X. Li,E. Considine, and M. Hallett. Transcranial direct current stimula-tion for the treatment of Parkinson’s disease. Journal of Neurology,Neurosurgery & Psychiatry, 81(10):1105–1111, 10 2010.[30] S. Bense, T. Stephan, T. A. Yousry, T. Brandt, and M. Dieterich.Multisensory Cortical Signal Increases and Decreases During Vestibu-lar Galvanic Stimulation (fMRI). J Neurophysiol, 85(2):886–899, 22001.[31] H. Bergman, A. Feingold, A. Nini, A. Raz, H. Slovin, M. Abeles, andE. Vaadia. Physiological aspects of information processing in the basalganglia of normal and parkinsonian primates. Trends in neurosciences,21(1):32–8, 1 1998.[32] J. A. Bernard, S. J. Peltier, B. L. Benson, J. L. Wiggins, S. M. Jaeggi,M. Buschkuehl, J. Jonides, C. S. Monk, and R. D. Seidler. DissociableFunctional Networks of the Human Dentate Nucleus. Cerebral Cortex,24(8):2151–2159, 8 2014.[33] M. Bikson, A. Name, and A. Rahman. Origins of specificity dur-ing tDCS: anatomical, activity-selective, and input-bias mechanisms.Frontiers in Human Neuroscience, 7:688, 2013.[34] W. Birkmayer and O. Hornykiewicz. The effect of l-3,4-dihydroxyphenylalanine (=DOPA) on akinesia in parkinsonism.Parkinsonism & related disorders, 4(2):59–60, 8 1998.[35] A. Bisecco, M. A. Rocca, E. Pagani, L. Mancini, C. Enzinger, A. Gallo,H. Vrenken, M. L. Stromillo, M. Copetti, D. L. Thomas, F. Fazekas,G. Tedeschi, F. Barkhof, N. D. Stefano, M. Filippi, and MAGNIMSNetwork. Connectivity-based parcellation of the thalamus in multiplesclerosis and its implications for cognitive impairment: A multicenterstudy. Human Brain Mapping, 36(7):2809–2825, 7 2015.132Bibliography[36] A. Bjo¨rklund and S. B. Dunnett. Dopamine neuron systems in thebrain: an update. Trends in Neurosciences, 30(5):194–202, 5 2007.[37] F. Blandini, G. Nappi, C. Tassorelli, and E. Martignoni. Functionalchanges of the basal ganglia circuitry in Parkinson’s disease. Progressin neurobiology, 62(1):63–88, 9 2000.[38] P. S. Boggio, R. Ferrucci, S. P. Rigonatti, P. Covre, M. Nitsche,A. Pascual-Leone, and F. Fregni. Effects of transcranial direct currentstimulation on working memory in patients with Parkinson’s disease.Journal of the Neurological Sciences, 249(1):31–38, 11 2006.[39] T. A. Boonstra, J. P. P. van Vugt, H. van der Kooij, and B. R.Bloem. Balance asymmetry in Parkinson’s disease and its contributionto freezing of gait. PloS one, 9(7):e102493, 2014.[40] J. L. W. Bosboom, D. Stoffers, C. J. Stam, B. W. van Dijk, J. Ver-bunt, H. W. Berendse, and E. C. Wolters. Resting state oscillatorybrain dynamics in Parkinson’s disease: an MEG study. Clinical neuro-physiology : official journal of the International Federation of ClinicalNeurophysiology, 117(11):2521–31, 12 2006.[41] C. Bosch-Bouju, B. I. Hyland, and L. C. Parr-Brownlie. Motor tha-lamus integration of cortical, cerebellar and basal ganglia informa-tion: implications for normal and parkinsonian conditions. Frontiersin Computational Neuroscience, 7:163, 2013.[42] G. Bottini, R. Sterzi, E. Paulesu, G. Vallar, S. F. Cappa, F. Erminio,R. E. Passingham, C. D. Frith, and R. S. Frackowiak. Identificationof the central vestibular projections in man: a positron emission to-mography activation study. Experimental brain research, 99(1):164–9,1994.[43] T. Brandt, M. Strupp, and M. Dieterich. Towards a concept of disor-ders of "higher vestibular function". Frontiers in integra-tive neuroscience, 8:47, 2014.[44] J.-S. Brittain and P. Brown. Oscillations and the basal ganglia: motorcontrol and beyond. NeuroImage, 85 Pt 2(Pt 2):637–47, 1 2014.[45] J.-S. Brittain, H. Cagnan, A. R. Mehta, T. A. Saifee, M. J. Edwards,and P. Brown. Distinguishing the central drive to tremor in Parkin-son’s disease and essential tremor. The Journal of neuroscience : theofficial journal of the Society for Neuroscience, 35(2):795–806, 1 2015.133Bibliography[46] D. Brooks and G. M. Halliday. Intralaminar nuclei of the thalamus inLewy body diseases. Brain Research Bulletin, 78(2-3):97–104, 2 2009.[47] P. Brown. Oscillatory nature of human basal ganglia activity: relation-ship to the pathophysiology of Parkinson’s disease. Movement disor-ders : official journal of the Movement Disorder Society, 18(4):357–63,4 2003.[48] P. Brown. Abnormal oscillatory synchronisation in the motor sys-tem leads to impaired movement. Current Opinion in Neurobiology,17(6):656–664, 12 2007.[49] P. Brown, P. Mazzone, A. Oliviero, M. G. Altibrandi, F. Pilato, P. A.Tonali, and V. Di Lazzaro. Effects of stimulation of the subthalamicarea on oscillatory pallidal activity in Parkinson’s disease. Experimen-tal Neurology, 188(2):480–490, 8 2004.[50] P. Brown, A. Oliviero, P. Mazzone, A. Insola, P. Tonali, and V. Di Laz-zaro. Dopamine dependency of oscillations between subthalamic nu-cleus and pallidum in Parkinson’s disease. The Journal of neuroscience: the official journal of the Society for Neuroscience, 21(3):1033–8, 22001.[51] P. Brown and D. Williams. Basal ganglia local field potential activity:character and functional significance in the human. Clinical neuro-physiology : official journal of the International Federation of ClinicalNeurophysiology, 116(11):2510–9, 11 2005.[52] E. N. Bruce, M. C. Bruce, and S. Vennelaganti. Sample entropy trackschanges in electroencephalogram power spectrum with sleep state andaging. Journal of clinical neurophysiology : official publication of theAmerican Electroencephalographic Society, 26(4):257–66, 8 2009.[53] A. Bruck, T. Kurki, V. Kaasinen, T. Vahlberg, and J. O. Rinne. Hip-pocampal and prefrontal atrophy in patients with early non-dementedParkinson’s disease is related to cognitive impairment. Journal ofNeurology, Neurosurgery & Psychiatry, 75(10):1467–1469, 10 2004.[54] R. Brun˜a, F. Maestu´, and E. Pereda. Phase locking value revisited:teaching new tricks to an old dog. Journal of neural engineering,15(5):056011, 10 2018.134Bibliography[55] S. F. Bucher, M. Dieterich, M. Wiesmann, A. Weiss, R. Zink, T. a.Yousry, and T. Brandt. Cerebral functional magnetic resonance imag-ing of vestibular, auditory, and nociceptive areas during galvanic stim-ulation. Annals of Neurology, 44(1):120–125, 7 1998.[56] G. Buzsaki and A. Draguhn. Neuronal Oscillations in Cortical Net-works. Science, 304(5679):1926–1929, 6 2004.[57] J. Cai, S. Lee, F. Ba, S. Garg, L. J. Kim, A. Liu, D. Kim, Z. J.Wang, and M. J. McKeown. Galvanic Vestibular Stimulation (GVS)Augments Deficient Pedunculopontine Nucleus (PPN) Connectivity inMild Parkinson’s Disease: fMRI Effects of Different Stimuli. Frontiersin neuroscience, 12:101, 2018.[58] P. Calabresi, V. Ghiglieri, P. Mazzocchetti, I. Corbelli, and B. Picconi.Levodopa-induced plasticity: a double-edged sword in Parkinson’s dis-ease? Philosophical Transactions of the Royal Society B: BiologicalSciences, 370(1672):20140184, 7 2015.[59] E. F. Cardoso, F. M. Maia, F. Fregni, M. L. Myczkowski, L. M. Melo,J. R. Sato, M. A. Marcolin, S. P. Rigonatti, A. C. Cruz, E. R. Barbosa,and E. Amaro. Depression in Parkinson’s disease: Convergence fromvoxel-based morphometry and functional magnetic resonance imagingin the limbic thalamus. NeuroImage, 47(2):467–472, 8 2009.[60] M. Cassidy, P. Mazzone, A. Oliviero, A. Insola, P. Tonali, V. Di Laz-zaro, and P. Brown. Movement-related changes in synchronizationin the human basal ganglia. Brain : a journal of neurology, 125(Pt6):1235–46, 6 2002.[61] M. Castle, M. S. Aymerich, C. Sanchez-Escobar, N. Gonzalo, J. A.Obeso, and J. L. Lanciego. Thalamic innervation of the direct andindirect basal ganglia pathways in the rat: Ipsi- and contralateralprojections. The Journal of Comparative Neurology, 483(2):143–153,3 2005.[62] A. S. Cauquil and B. L. Day. Galvanic vestibular stimulation modu-lates voluntary movement of the human upper body. The Journal ofphysiology, 513 ( Pt 2:611–9, 12 1998.[63] J. F. Cavanagh, L. Zambrano-Vazquez, and J. J. B. Allen. Thetalingua franca: A common mid-frontal substrate for action monitoringprocesses. Psychophysiology, 49(2):220–238, 2 2012.135Bibliography[64] X. Chen, A. Liu, Q. Chen, Y. Liu, L. Zou, and M. J. McKeown.Simultaneous ocular and muscle artifact removal from EEG data byexploiting diverse statistics. Computers in Biology and Medicine, 88:1–10, 9 2017.[65] X. Chen, A. Liu, J. Chiang, Z. J. Wang, M. J. McKeown, and R. K.Ward. Removing Muscle Artifacts From EEG Data: Multichannel orSingle-Channel Techniques? IEEE Sensors Journal, 16(7):1986–1997,4 2016.[66] X. Chen, H. Peng, F. Yu, and K. Wang. Independent Vector AnalysisApplied to Remove Muscle Artifacts in EEG Data. IEEE Transactionson Instrumentation and Measurement, 66(7):1770–1779, 7 2017.[67] X. Chen, Z. J. Wang, and M. McKeown. Joint Blind Source Separa-tion for Neurophysiological Data Analysis: Multiset and multimodalmethods. IEEE Signal Processing Magazine, 33(3):86–107, 5 2016.[68] S. J. Chung, J. J. Lee, J. H. Ham, B. S. Ye, P. H. Lee, and Y. H. Sohn.Striatal Dopamine Depletion Patterns and Early Non-Motor Burdenin Parkinsons Disease. PloS one, 11(8):e0161316, 2016.[69] D. O. Claassen, K. E. McDonell, M. Donahue, S. Rawal, S. A. Wylie,J. S. Neimat, H. Kang, P. Hedera, D. Zald, B. Landman, B. Dawant,and S. Rane. Cortical asymmetry in Parkinson’s disease: early sus-ceptibility of the left hemisphere. Brain and behavior, 6(12):e00573,2016.[70] L. Clemmensen, T. Hastie, D. Witten, and B. Ersbøll. Sparse Dis-criminant Analysis. Technometrics, 53(4):406–413, 11 2011.[71] P. Cordo, J. T. Inglis, S. Verschueren, J. J. Collins, D. M. Merfeld,S. Rosenblum, S. Buckley, and F. Moss. Noise in human muscle spin-dles. Nature, 383(6603):769–770, 10 1996.[72] N. Correa, T. Adali, Y.-O. Li, and V. Calhoun. Canonical Corre-lation Analysis for Data Fusion and Group Inferences. IEEE SignalProcessing Magazine, 27(4):39–50, 2010.[73] G. C. Cotzias, P. S. Papavasiliou, and R. Gellene. Modification ofParkinsonism Chronic Treatment with L-Dopa. New England Journalof Medicine, 280(7):337–345, 2 1969.136Bibliography[74] R. C. Craddock, G. James, P. E. Holtzheimer, X. P. Hu, and H. S.Mayberg. A whole brain fMRI atlas generated via spatially constrainedspectral clustering. Human Brain Mapping, 33(8):1914–1928, 8 2012.[75] K. E. Cullen. The vestibular system: multimodal integration andencoding of self-motion for motor control. Trends in neurosciences,35(3):185–96, 3 2012.[76] R. G. Cury, V. Fraix, A. Castrioto, M. A. Pe´rez Ferna´ndez, P. Krack,S. Chabardes, E. Seigneuret, E. J. L. Alho, A.-L. Benabid, andE. Moro. Thalamic deep brain stimulation for tremor in Parkinsondisease, essential tremor, and dystonia. Neurology, 89(13):1416–1423,9 2017.[77] D. Cyron. Mental Side Effects of Deep Brain Stimulation (DBS) forMovement Disorders: The Futility of Denial. Frontiers in integrativeneuroscience, 10:17, 2016.[78] N. Dahodwala, A. Siderowf, M. Xie, E. Noll, M. Stern, and D. S.Mandell. Racial differences in the diagnosis of Parkinson’s disease.Movement disorders : official journal of the Movement Disorder Soci-ety, 24(8):1200–5, 6 2009.[79] C. A. Davie. A review of Parkinson’s disease. British medical bulletin,86(1):109–27, 1 2008.[80] B. L. Day, A. Se´verac Cauquil, L. Bartolomei, M. A. Pastor, and I. N.Lyon. Human body-segment tilts induced by galvanic stimulation:a vestibularly driven balance protection mechanism. The Journal ofphysiology, 500 ( Pt 3:661–72, 5 1997.[81] B. L. Day, A. Se´verac Cauquil, L. Bartolomei, M. A. Pastor, and I. N.Lyon. Human body-segment tilts induced by galvanic stimulation:a vestibularly driven balance protection mechanism. The Journal ofphysiology, 500 ( Pt 3:661–72, 5 1997.[82] L. Defebvre, J. L. Bourriez, K. Dujardin, P. Derambure, A. Deste´e, andJ. D. Guieu. Spatiotemporal study of Bereitschaftspotential and event-related desynchronization during voluntary movement in Parkinson’sdisease. Brain topography, 6(3):237–44, 1994.[83] M. DeLong and T. Wichmann. Changing views of basal ganglia circuitsand circuit disorders. Clinical EEG and neuroscience, 41(2):61–7, 42010.137Bibliography[84] A. Delorme and S. Makeig. EEGLAB: an open source toolbox foranalysis of single-trial EEG dynamics including independent compo-nent analysis. Journal of Neuroscience Methods, 134(1):9–21, 3 2004.[85] A. Delorme, J. Palmer, R. Oostenveld, J. Onton, and S. Makeig. Com-paring Results of Algorithms Implementing Blind Source Separationof EEG Data. unpublished.[86] A. Delorme, T. Sejnowski, and S. Makeig. Enhanced detection ofartifacts in EEG data using higher-order statistics and independentcomponent analysis. NeuroImage, 34(4):1443–9, 2 2007.[87] Z.-D. Deng, S. H. Lisanby, and A. V. Peterchev. Electric field depthfo-cality tradeoff in transcranial magnetic stimulation: Simulation com-parison of 50 coil designs. Brain Stimulation, 6(1):1–13, 1 2013.[88] M. Desmurget and A. Sirigu. A parietal-premotor network for move-ment intention and motor awareness. Trends in Cognitive Sciences,13(10):411–419, 10 2009.[89] G. Deuschl, C. Schade-Brittinger, P. Krack, J. Volkmann, H. Scha¨fer,K. Bo¨tzel, C. Daniels, A. Deutschla¨nder, U. Dillmann, W. Eisner,D. Gruber, W. Hamel, J. Herzog, R. Hilker, S. Klebe, M. Kloß, J. Koy,M. Krause, A. Kupsch, D. Lorenz, S. Lorenzl, H. M. Mehdorn, J. R.Moringlane, W. Oertel, M. O. Pinsker, H. Reichmann, A. Reuß, G.-H. Schneider, A. Schnitzler, U. Steude, V. Sturm, L. Timmermann,V. Tronnier, T. Trottenberg, L. Wojtecki, E. Wolf, W. Poewe, J. Vo-ges, and N. S. German Parkinson Study Group. A Randomized Trial ofDeep-Brain Stimulation for Parkinson’s Disease. New England Journalof Medicine, 355(9):896–908, 8 2006.[90] D. Devos, E. Labyt, P. Derambure, J. L. Bourriez, F. Cassim, J. D.Guieu, A. Deste´e, and L. Defebvre. Effect of L-Dopa on the patternof movement-related (de)synchronisation in advanced Parkinson’s dis-ease. Neurophysiologie clinique = Clinical neurophysiology, 33(5):203–12, 11 2003.[91] J. P. Dmochowski, A. Datta, M. Bikson, Y. Su, and L. C. Parra.Optimized multi-electrode stimulation increases focality and intensityat target. Journal of Neural Engineering, 8(4):046011, 8 2011.138Bibliography[92] D. Doruk, Z. Gray, G. L. Bravo, A. Pascual-Leone, and F. Fregni.Effects of tDCS on executive function in Parkinson’s disease. Neuro-science Letters, 582:27–31, 10 2014.[93] L. M. F. Doyle, A. A. Ku¨hn, M. Hariz, A. Kupsch, G.-H. Schneider,and P. Brown. Levodopa-induced modulation of subthalamic betaoscillations during self-paced movements in patients with Parkinson’sdisease. European Journal of Neuroscience, 21(5):1403–1412, 3 2005.[94] S. du Lac and S. G. Lisberger. Cellular processing of temporal informa-tion in medial vestibular nucleus neurons. The Journal of neuroscience: the official journal of the Society for Neuroscience, 15(12):8000–10,12 1995.[95] K. T. E. O. Dubbelink, M. M. Schoonheim., J. B. Deijen, J. W. R.Twisk, F. Barkhof, H. W. Berendse, K. T. E. Olde Dubbelink, M. M.Schoonheim, J. B. Deijen, J. W. R. Twisk, F. Barkhof, and H. W.Berendse. Functional connectivity and cognitive decline over 3 yearsin Parkinson disease. Neurology, 83(22):2046–2053, 11 2014.[96] R. O. Duda, P. E. Hart, and D. G. Stork. Pattern Classification, 2001.[97] A. K. Engel and P. Fries. Beta-band oscillations signalling the statusquo? Current Opinion in Neurobiology, 20(2):156–165, 4 2010.[98] D. Erskine, A. J. Thomas, J. Attems, J.-P. Taylor, I. G. McKeith,C. M. Morris, and A. A. Khundakar. Specific patterns of neuronalloss in the pulvinar nucleus in dementia with lewy bodies. MovementDisorders, 32(3):414–422, 3 2017.[99] S. Espenhahn, A. O. de Berker, B. C. M. van Wijk, H. E. Rossiter,and N. S. Ward. Movement-related beta oscillations show high intra-individual reliability. NeuroImage, 147:175–185, 2017.[100] A. Eusebio, A. Pogosyan, S. Wang, B. Averbeck, L. D. Gaynor, S. Can-tiniaux, T. Witjas, P. Limousin, J.-P. Azulay, and P. Brown. Reso-nance in subthalamo-cortical circuits in Parkinson’s disease. Brain,132(8):2139–2150, 8 2009.[101] E. Faught and W. Tatum. Trigeminal stimulation: A superhighwayto the brain? Neurology, 80(9):780–781, 2 2013.139Bibliography[102] J. M. Fearnley and A. J. Lees. Ageing and Parkinson’s disease: sub-stantia nigra regional selectivity. Brain : a journal of neurology, 114( Pt 5:2283–301, 10 1991.[103] J. Fell and N. Axmacher. The role of phase synchronization in memoryprocesses. Nature Reviews Neuroscience, 12(2):105–118, 2 2011.[104] S. J. Fellows and J. Noth. Grip force abnormalities in de novo Parkin-son’s disease. Movement Disorders, 19(5):560–565, 5 2004.[105] R. Ferrucci, F. Cortese, M. Bianchi, D. Pittera, R. Turrone, T. Bocci,B. Borroni, M. Vergari, F. Cogiamanian, G. Ardolino, A. Di Fonzo,A. Padovani, and A. Priori. Cerebellar and Motor Cortical Transcra-nial Stimulation Decrease Levodopa-Induced Dyskinesias in Parkin-sons Disease. The Cerebellum, 15(1):43–47, 2 2016.[106] A. Fertonani and C. Miniussi. Transcranial Electrical Stimulation:What We Know and Do Not Know About Mechanisms. The Neuro-scientist, 23(2):109–123, 4 2017.[107] M. Feurra, G. Bianco, E. Santarnecchi, M. Del Testa, A. Rossi, andS. Rossi. Frequency-Dependent Tuning of the Human Motor SystemInduced by Transcranial Oscillatory Potentials. Journal of Neuro-science, 31(34):12165–12170, 8 2011.[108] P. Filzmoser and V. Todorov. Review of robust multivariate statisticalmethods in high dimension. Analytica Chimica Acta, 705(1-2):2–14,10 2011.[109] S. D. Fisher and J. N. J. Reynolds. The intralaminar thalamus-anexpressway linking visual stimuli to circuits determining agency andaction selection. Frontiers in Behavioral Neuroscience, 8:115, 4 2014.[110] R. Fitzpatrick, D. Burke, and S. C. Gandevia. Task-dependent re-flex responses and movement illusions evoked by galvanic vestibularstimulation in standing humans. The Journal of physiology, 478 ( Pt2:363–72, 7 1994.[111] R. C. Fitzpatrick and B. L. Day. Probing the human vestibular systemwith galvanic stimulation. Journal of applied physiology, 96(6):2301–16, 6 2004.140Bibliography[112] G. Foffani, G. Ardolino, B. Meda, M. Egidi, P. Rampini, E. Caputo,G. Baselli, and A. Priori. Altered subthalamo-pallidal synchronisationin parkinsonian dyskinesias. Journal of Neurology, Neurosurgery &Psychiatry, 76(3):426–428, 3 2005.[113] P. A. Forbes, C. J. Dakin, A. M. Geers, M. P. Vlaar, R. Happee,G. P. Siegmund, A. C. Schouten, and J.-S. Blouin. Electrical Vestibu-lar Stimuli to Enhance Vestibulo-Motor Output and Improve SubjectComfort. PLoS ONE, 9(1):e84385, 2014.[114] F. Fregni, P. S. Boggio, M. C. Santos, M. Lima, A. L. Vieira,S. P. Rigonatti, M. T. A. Silva, E. R. Barbosa, M. A. Nitsche, andA. Pascual-Leone. Noninvasive cortical stimulation with transcranialdirect current stimulation in Parkinson’s disease. Movement Disor-ders, 21(10):1693–1702, 10 2006.[115] F. Fregni and A. Pascual-Leone. Technology Insight: noninvasive brainstimulation in neurologyperspectives on the therapeutic potential ofrTMS and tDCS. Nature Clinical Practice Neurology, 3(7):383–393, 72007.[116] P. Fries. Neuronal Gamma-Band Synchronization as a FundamentalProcess in Cortical Computation. Annual Review of Neuroscience,32(1):209–224, 6 2009.[117] E. C. Fuchs, H. Doheny, H. Faulkner, A. Caputi, R. D. Traub, A. Bib-big, N. Kopell, M. A. Whittington, and H. Monyer. Genetically alteredAMPA-type glutamate receptor kinetics in interneurons disrupt long-range synchrony of gamma oscillation. Proceedings of the NationalAcademy of Sciences, 98(6):3571–3576, 3 2001.[118] R. Fuentes, P. Petersson, W. B. Siesser, M. G. Caron, and M. A. L.Nicolelis. Spinal Cord Stimulation Restores Locomotion in AnimalModels of Parkinson’s Disease. Science, 323(5921):1578–1582, 3 2009.[119] R. Galambos and S. Makeig. Physiological studies of central maskingin man. I: The effects of noise on the 40-Hz steady-state response. TheJournal of the Acoustical Society of America, 92(5):2683–90, 11 1992.[120] M. N. Gallay, D. Jeanmonod, J. Liu, and A. Morel. Human pallidotha-lamic and cerebellothalamic tracts: anatomical basis for functionalstereotactic neurosurgery. Brain structure & function, 212(6):443–63,8 2008.141Bibliography[121] A. Galvan and T. Wichmann. Pathophysiology of parkinsonism. Clin-ical neurophysiology : official journal of the International Federationof Clinical Neurophysiology, 119(7):1459–74, 7 2008.[122] A. Garg, S. Appel-Cresswell, K. Popuri, M. J. McKeown, and M. F.Beg. Morphological alterations in the caudate, putamen, pallidum,and thalamus in Parkinson’s disease. Frontiers in neuroscience, 9:101,2015.[123] D. D. Garrett, G. R. Samanez-Larkin, S. W. MacDonald, U. Linden-berger, A. R. McIntosh, and C. L. Grady. Moment-to-moment brainsignal variability: A next frontier in human brain mapping? Neuro-science & Biobehavioral Reviews, 37(4):610–624, 5 2013.[124] J. S. George, J. Strunk, R. Mak-McCully, M. Houser, H. Poizner, andA. R. Aron. Dopaminergic therapy in Parkinson’s disease decreasescortical beta band coherence in the resting state and increases corti-cal beta band power during executive control. NeuroImage: Clinical,3:261–270, 2013.[125] J. Ghika, J.-G. Villemure, H. Fankhauser, J. Favre, G. Assal, andF. Ghika-Schmid. Efficiency and safety of bilateral contemporaneouspallidal stimulation (deep brain stimulation) in levodopa-responsivepatients with Parkinson’s disease with severe motor fluctuations: a2-year follow-up review. Journal of Neurosurgery, 89(5):713–718, 111998.[126] G. Giannicola, S. Marceglia, L. Rossi, S. Mrakic-Sposta, P. Rampini,F. Tamma, F. Cogiamanian, S. Barbieri, and A. Priori. The effects oflevodopa and ongoing deep brain stimulation on subthalamic beta os-cillations in Parkinson’s disease. Experimental Neurology, 226(1):120–127, 11 2010.[127] T. Gilbertson, E. Lalo, L. Doyle, V. Di Lazzaro, B. Cioni, andP. Brown. Existing Motor State Is Favored at the Expense of NewMovement during 13-35 Hz Oscillatory Synchrony in the Human Cor-ticospinal System. Journal of Neuroscience, 25(34):7771–7779, 8 2005.[128] J. M. Goldberg, C. E. Smith, and C. Fernandez. Relation betweendischarge regularity and responses to externally applied galvanic cur-rents in vestibular nerve afferents of the squirrel monkey. Journal ofNeurophysiology, 51(6):1236–1256, 6 1984.142Bibliography[129] A. M. Gordon, P. E. Ingvarsson, and H. Forssberg. Anticipatory Con-trol of Manipulative Forces in Parkinson’s Disease. Experimental Neu-rology, 145(2):477–488, 6 1997.[130] L. T. Grinberg, U. Rueb, A. T. d. L. Alho, and H. Heinsen. Brainstempathology and non-motor symptoms in PD. Journal of the Neurolog-ical Sciences, 289(1-2):81–88, 2 2010.[131] S. J. Groiss, L. Wojtecki, M. Su¨dmeyer, and A. Schnitzler. Deep brainstimulation in Parkinson’s disease. Therapeutic advances in neurolog-ical disorders, 2(6):20–8, 11 2009.[132] D. M. Groppe, S. Bickel, C. J. Keller, S. K. Jain, S. T. Hwang,C. Harden, and A. D. Mehta. Dominant frequencies of resting humanbrain activity as measured by the electrocorticogram. NeuroImage,79:223–33, 10 2013.[133] R. P. Guimara˜es, M. C. Arci Santos, A. Dagher, L. S. Cam-pos, P. Azevedo, L. G. Piovesana, B. M. De Campos, K. Larcher,Y. Zeighami, A. C. Scarparo Amato-Filho, F. Cendes, and A. C. F.D’Abreu. Pattern of Reduced Functional Connectivity and StructuralAbnormalities in Parkinson’s Disease: An Exploratory Study. Fron-tiers in neurology, 7:243, 2016.[134] C. Gurvich, J. J. Maller, B. Lithgow, S. Haghgooie, and J. Kulkarni.Vestibular insights into cognition and psychiatry. Brain Research,1537:244–259, 11 2013.[135] S. Haber and N. R. Mcfarland. The Place of the Thalamus in FrontalCortical-Basal Ganglia Circuits. The Neuroscientist, 7(4):315–324, 82001.[136] S. N. Haber and R. Calzavara. The cortico-basal ganglia integrativenetwork: The role of the thalamus. Brain Research Bulletin, 78(2-3):69–74, 2 2009.[137] M. Haghighat, M. Abdel-Mottaleb, and W. Alhalabi. DiscriminantCorrelation Analysis: Real-Time Feature Level Fusion for MultimodalBiometric Recognition. IEEE Transactions on Information Forensicsand Security, 11(9):1984–1996, 9 2016.[138] N. Hajos, J. Pa´lhalmi, E. O. Mann, B. Ne´meth, O. Paulsen, and T. F.Freund. Spike Timing of Distinct Types of GABAergic Interneuron143Bibliographyduring Hippocampal Gamma Oscillations In Vitro. Journal of Neuro-science, 24(41):9127–9137, 10 2004.[139] G. M. Halliday. Thalamic changes in Parkinson’s disease. Parkinson-ism & Related Disorders, 15:S152–S155, 12 2009.[140] G. M. Halliday, V. Macdonald, and J. M. Henderson. A comparisonof degeneration in motor thalamus and cortex between progressivesupranuclear palsy and Parkinson’s disease. Brain, 128(10):2272–2280,10 2005.[141] S. Hanslmayr, W. Klimesch, P. Sauseng, W. Gruber, M. Doppelmayr,R. Freunberger, and T. Pecherstorfer. Visual discrimination perfor-mance is related to decreased alpha amplitude but increased phaselocking. Neuroscience letters, 375(1):64–8, 2 2005.[142] R. Hari, R. Salmelin, J. P. Ma¨kela¨, S. Salenius, and M. Helle. Mag-netoencephalographic cortical rhythms. International journal of psy-chophysiology : official journal of the International Organization ofPsychophysiology, 26(1-3):51–62, 6 1997.[143] S.-b. He, C.-y. Liu, L.-d. Chen, Z.-n. Ye, Y.-p. Zhang, W.-g. Tang,B.-d. Wang, and X. Gao. Meta-Analysis of Visual Evoked Potentialand Parkinsons Disease. Parkinson’s Disease, 2018:1–8, 7 2018.[144] X. He, Y. Zhang, J. Chen, C. Xie, R. Gan, R. Yang, L. Wang, K. Nie,and L. Wang. The patterns of EEG changes in early-onset Parkinson’sdisease patients. International Journal of Neuroscience, 127(11):1028–1035, 11 2017.[145] E. Heinrichs-Graham and T. W. Wilson. Is an absolute level of corticalbeta suppression required for proper movement? Magnetoencephalo-graphic evidence from healthy aging. NeuroImage, 134:514–521, 72016.[146] E. Heinrichs-Graham, T. W. Wilson, P. M. Santamaria, S. K. Hei-thoff, D. Torres-Russotto, J. A. L. Hutter-Saunders, K. A. Estes, J. L.Meza, R. L. Mosley, and H. E. Gendelman. Neuromagnetic evidenceof abnormal movement-related beta desynchronization in Parkinson’sdisease. Cerebral cortex (New York, N.Y. : 1991), 24(10):2669–78, 102014.144Bibliography[147] R. F. Helfrich, T. R. Schneider, S. Rach, S. A. Trautmann-Lengsfeld,A. K. Engel, and C. S. Herrmann. Entrainment of brain oscillationsby transcranial alternating current stimulation. Current biology : CB,24(3):333–9, 2 2014.[148] R. C. Helmich, L. C. Derikx, M. Bakker, R. Scheeringa, B. R. Bloem,and I. Toni. Spatial Remapping of Cortico-striatal Connectivity inParkinson’s Disease. Cerebral Cortex, 20(5):1175–1186, 5 2010.[149] J. M. Henderson, K. Carpenter, H. Cartwright, and G. M. Halliday.Loss of thalamic intralaminar nuclei in progressive supranuclear palsyand Parkinson’s disease: clinical and therapeutic implications. Brain: a journal of neurology, 123 ( Pt 7:1410–21, 7 2000.[150] C. S. Herrmann. Human EEG responses to 1-100 Hz flicker: resonancephenomena in visual cortex and their potential correlation to cognitivephenomena. Experimental Brain Research, 137(3-4):346–353, 4 2001.[151] C. S. Herrmann, S. Rach, T. Neuling, and D. Stru¨ber. Transcranial al-ternating current stimulation: a review of the underlying mechanismsand modulation of cognitive processes. Frontiers in Human Neuro-science, 7:279, 2013.[152] L. Hirsch, N. Jette, A. Frolkis, T. Steeves, and T. Pringsheim. TheIncidence of Parkinson’s Disease: A Systematic Review and Meta-Analysis. Neuroepidemiology, 46(4):292–300, 2016.[153] S. W. Hughes and V. Crunelli. Thalamic mechanisms of EEG al-pha rhythms and their pathological implications. The Neuroscientist: a review journal bringing neurobiology, neurology and psychiatry,11(4):357–72, 8 2005.[154] F. Hummel, P. Celnik, P. Giraux, A. Floel, W.-H. Wu, C. Gerloff, andL. G. Cohen. Effects of non-invasive cortical stimulation on skilledmotor function in chronic stroke. Brain, 128(3):490–499, 1 2005.[155] W. D. Hutchison, J. O. Dostrovsky, J. R. Walters, R. Courtemanche,T. Boraud, J. Goldberg, and P. Brown. Neuronal Oscillations in theBasal Ganglia and Movement Disorders: Evidence from Whole Animaland Human Recordings. Journal of Neuroscience, 24(42):9240–9243,10 2004.145Bibliography[156] K. Hwang, M. A. Bertolero, W. B. Liu, and M. D’Esposito. The Hu-man Thalamus Is an Integrative Hub for Functional Brain Networks.The Journal of neuroscience : the official journal of the Society forNeuroscience, 37(23):5594–5607, 2017.[157] I. A. Ilinsky, K. Kultas-Ilinsky, A. Rosina, and M. Haddy. Quantitativeevaluation of crossed and uncrossed projections from basal ganglia andcerebellum to the cat thalamus. Neuroscience, 21(1):207–27, 4 1987.[158] M. Inase, J. A. Buford, and M. E. Anderson. Changes in the control ofarm position, movement, and thalamic discharge during local inactiva-tion in the globus pallidus of the monkey. Journal of Neurophysiology,75(3):1087–1104, 3 1996.[159] C. James, A. Stathis, and V. G. Macefield. Vestibular and pulse-related modulation of skin sympathetic nerve activity during sinu-soidal galvanic vestibular stimulation in human subjects. Experimentalbrain research, 202(2):291–8, 4 2010.[160] J. Jankovic. Parkinson’s disease: clinical features and diagnosis. Jour-nal of Neurology, Neurosurgery & Psychiatry, 79(4):368–376, 4 2008.[161] H. Jasper and W. Penfield. Electrocorticograms in man: Effect ofvoluntary movement upon the electrical activity of the precentralgyrus. Archiv fur Psychiatrie und Nervenkrankheiten, 183(1-2):163–174, 1949.[162] N. Jenkinson and P. Brown. New insights into the relationship betweendopamine, beta oscillations and motor function. Trends in Neuro-sciences, 34(12):611–618, 12 2011.[163] J. Jeong. EEG dynamics in patients with Alzheimer’s disease. Clinicalneurophysiology : official journal of the International Federation ofClinical Neurophysiology, 115(7):1490–505, 7 2004.[164] B. Ji, Z. Li, K. Li, L. Li, J. Langley, H. Shen, S. Nie, R. Zhang, andX. Hu. Dynamic thalamus parcellation from resting-state fMRI data.Human Brain Mapping, 37(3):954–967, 3 2016.[165] W. Jian, M. Chen, and D. J. McFarland. EEG Based Zero-phasePhase-locking Value (PLV) and Effects of Spatial Filtering DuringActual Movement. Brain research bulletin, 130:156, 2017.146Bibliography[166] Jianbo Shi and J. Malik. Normalized cuts and image segmentation.IEEE Transactions on Pattern Analysis and Machine Intelligence,22(8):888–905, 2000.[167] S. M. Jones, T. A. Jones, K. N. Mills, and G. C. Gaines. Anatom-ical and Physiological Considerations in Vestibular Dysfunction andCompensation. Seminars in hearing, 30(4):231–241, 2009.[168] N. Jordan, H. J. Sagar, and J. A. Cooper. Cognitive components of re-action time in Parkinson’s disease. Journal of neurology, neurosurgery,and psychiatry, 55(8):658–64, 8 1992.[169] R. A. Joundi, N. Jenkinson, J.-S. Brittain, T. Z. Aziz, and P. Brown.Driving oscillatory activity in the human cortex enhances motor per-formance. Current Biology, 22(5):403–407, 3 2012.[170] T. Jubault, S. M. Brambati, C. Degroot, B. Kullmann, A. P. Strafella,A.-L. Lafontaine, S. Chouinard, and O. Monchi. Regional brain stematrophy in idiopathic Parkinson’s disease detected by anatomical MRI.PloS one, 4(12):e8247, 12 2009.[171] H. Kataoka, Y. Okada, T. Kiriyama, Y. Kita, J. Nakamura,S. Morioka, K. Shomoto, and S. Ueno. Can Postural Instability Re-spond to Galvanic Vestibular Stimulation in Patients with ParkinsonsDisease? Journal of Movement Disorders, 9(1):40–43, 12 2016.[172] J. R. Kettenring. Canonical Analysis of Several Sets of Variables.Biometrika, 58(3):433, 12 1971.[173] S. Khan and R. Chang. Anatomy of the vestibular system: a review.NeuroRehabilitation, 32(3):437–43, 2013.[174] M. Khoshnam, D. M. C. Ha¨ner, E. Kuatsjah, X. Zhang, and C. Menon.Effects of Galvanic Vestibular Stimulation on Upper and Lower Ex-tremities Motor Symptoms in Parkinsons Disease. Frontiers in Neu-roscience, 12:633, 9 2018.[175] B. E. Kilavik, M. Zaepffel, A. Brovelli, W. A. MacKay, and A. Riehle.The ups and downs of beta oscillations in sensorimotor cortex. Exper-imental Neurology, 245:15–26, 7 2013.[176] D. Kim, B. Park, and H. Park. Functional connectivity-based identi-fication of subdivisions of the basal ganglia and thalamus using mul-147Bibliographytilevel independent component analysis of resting state fMRI. HumanBrain Mapping, 34(6):1371–1385, 6 2013.[177] D. J. Kim, V. Yogendrakumar, J. Chiang, E. Ty, Z. J. Wang, andM. J. McKeown. Noisy Galvanic Vestibular Stimulation Modulatesthe Amplitude of EEG Synchrony Patterns. PLoS ONE, 8(7):e69055,7 2013.[178] N. Kim, J. W. Barter, T. Sukharnikova, and H. H. Yin. Striatal firingrate reflects head movement velocity. European Journal of Neuro-science, 40(10):3481–3490, 11 2014.[179] T. Kim, T. Eltoft, and T.-W. Lee. Independent Vector Analysis:An Extension of ICA to Multivariate Components. pages 165–172.Springer, Berlin, Heidelberg, 2006.[180] K. Kitajo, S. M. Doesburg, K. Yamanaka, D. Nozaki, L. M. Ward,and Y. Yamamoto. Noise-induced large-scale phase synchronizationof human-brain activity associated with behavioural stochastic reso-nance. Europhysics Letters (EPL), 80(4):40009, 11 2007.[181] G. Kleiner-Fisman, J. Herzog, D. N. Fisman, F. Tamma, K. E. Lyons,R. Pahwa, A. E. Lang, and G. Deuschl. Subthalamic nucleus deepbrain stimulation: Summary and meta-analysis of outcomes. Move-ment Disorders, 21(S14):S290–S304, 6 2006.[182] W. Klimesch. EEG alpha and theta oscillations reflect cognitive andmemory performance: a review and analysis. Brain research. Brainresearch reviews, 29(2-3):169–95, 4 1999.[183] W. Klimesch, M. Doppelmayr, A. Yonelinas, N. E. Kroll, M. Lazzara,D. Ro¨hm, and W. Gruber. Theta synchronization during episodicretrieval: neural correlates of conscious awareness. Brain research.Cognitive brain research, 12(1):33–8, 8 2001.[184] W. Klimesch, R. Freunberger, P. Sauseng, and W. Gruber. A short re-view of slow phase synchronization and memory: Evidence for controlprocesses in different memory systems? Brain Research, 1235:31–44,10 2008.[185] W. Klimesch, P. Sauseng, and S. Hanslmayr. EEG alpha oscillations:The inhibitiontiming hypothesis. Brain Research Reviews, 53(1):63–88, 1 2007.148Bibliography[186] W. Klomjai, R. Katz, and A. Lackmy-Valle´e. Basic principles of tran-scranial magnetic stimulation (TMS) and repetitive TMS (rTMS). An-nals of Physical and Rehabilitation Medicine, 58(4):208–213, 9 2015.[187] S. Kohli and A. J. Casson. Removal of Transcranial a.c. Current Stim-ulation artifact from simultaneous EEG recordings by superpositionof moving averages. Conference proceedings : ... Annual InternationalConference of the IEEE Engineering in Medicine and Biology Society.IEEE Engineering in Medicine and Biology Society. Annual Confer-ence, 2015:3436–9, 1 2015.[188] P. Krack, A. Batir, N. Van Blercom, S. Chabardes, V. Fraix, C. Ar-douin, A. Koudsie, P. D. Limousin, A. Benazzouz, J. F. LeBas, A.-L.Benabid, and P. Pollak. Five-Year Follow-up of Bilateral Stimulationof the Subthalamic Nucleus in Advanced Parkinson’s Disease. NewEngland Journal of Medicine, 349(20):1925–1934, 11 2003.[189] V. Krause, C. Wach, M. Su¨dmeyer, S. Ferrea, A. Schnitzler, andB. Pollok. Cortico-muscular coupling and motor performance are mod-ulated by 20 Hz transcranial alternating current stimulation (tACS)in Parkinson’s disease. Frontiers in human neuroscience, 7:928, 2013.[190] A. C. Kreitzer and R. C. Malenka. Endocannabinoid-mediated res-cue of striatal LTD and motor deficits in Parkinson’s disease models.Nature, 445(7128):643–647, 2 2007.[191] A. A. Ku¨hn, A. Tsui, T. Aziz, N. Ray, C. Bru¨cke, A. Kupsch, G.-H.Schneider, and P. Brown. Pathological synchronisation in the sub-thalamic nucleus of patients with Parkinson’s disease relates to bothbradykinesia and rigidity. Experimental Neurology, 215(2):380–387, 22009.[192] V. Kumar, S. Mang, and W. Grodd. Direct diffusion-based par-cellation of the human thalamus. Brain Structure and Function,220(3):1619–1635, 5 2015.[193] Y. Kumar, M. L. Dewal, and R. S. Anand. Features extraction ofEEG signals using approximate and sample entropy. In 2012 IEEEStudents’ Conference on Electrical, Electronics and Computer Science,pages 1–5. IEEE, 3 2012.149Bibliography[194] K. Kurata. Activity Properties and Location of Neurons in the Mo-tor Thalamus That Project to the Cortical Motor Areas in Monkeys.Journal of Neurophysiology, 94(1):550–566, 7 2005.[195] J. P. Lachaux, E. Rodriguez, J. Martinerie, and F. J. Varela. Measur-ing phase synchrony in brain signals. Human brain mapping, 8(4):194–208, 1999.[196] H. Lai, T. Tsumori, T. Shiroyama, S. Yokota, K. Nakano, and Y. Ya-sui. Morphological evidence for a vestibulo-thalamo-striatal pathwayvia the parafascicular nucleus in the rat. Brain research, 872(1-2):208–14, 7 2000.[197] P. Lakatos, M. N. O’Connell, and A. Barczak. Pondering the Pulvinar.Neuron, 89(1):5–7, 1 2016.[198] D. E. Lake, J. S. Richman, M. P. Griffin, and J. R. Moorman. Sampleentropy analysis of neonatal heart rate variability. American Jour-nal of Physiology-Regulatory, Integrative and Comparative Physiology,283(3):R789–R797, 9 2002.[199] J. L. Lanciego, N. Luquin, and J. A. Obeso. Functional neuroanatomyof the basal ganglia. Cold Spring Harbor perspectives in medicine,2(12):a009621, 12 2012.[200] Lang Tong, Guanghan Xu, and T. Kailath. Blind identification andequalization based on second-order statistics: a time domain approach.IEEE Transactions on Information Theory, 40(2):340–349, 3 1994.[201] T. Langheinrich, L. Tebartz van Elst, W. A. Lagre`ze, M. Bach, C. H.Lu¨cking, and M. W. Greenlee. Visual contrast response functionsin Parkinson’s disease: evidence from electroretinograms, visuallyevoked potentials and psychophysics. Clinical neurophysiology : of-ficial journal of the International Federation of Clinical Neurophysiol-ogy, 111(1):66–74, 1 2000.[202] H. Laufs, A. Kleinschmidt, A. Beyerle, E. Eger, A. Salek-Haddadi,C. Preibisch, and K. Krakow. EEG-correlated fMRI of human alphaactivity. NeuroImage, 19(4):1463–76, 8 2003.[203] H. J. Lee, A. J. Weitz, D. Bernal-Casas, B. A. Duffy, M. Choy, A. V.Kravitz, A. C. Kreitzer, and J. H. Lee. Activation of Direct andIndirect Pathway Medium Spiny Neurons Drives Distinct Brain-wideResponses. Neuron, 91(2):412–424, 7 2016.150Bibliography[204] S. Lee, D. Kim, and M. J. McKeown. Galvanic Vestibular Stimulation(GVS) effects on impaired interhemispheric connectivity in Parkin-son’s Disease. In 2017 39th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC), volume2017, pages 2109–2113. IEEE, 7 2017.[205] S. Lee, D. J. Kim, D. Svenkeson, G. Parras, M. M. K. Oishi, and M. J.McKeown. Multifaceted effects of noisy galvanic vestibular stimula-tion on manual tracking behavior in Parkinson’s disease. Frontiers inSystems Neuroscience, 9:5, 2015.[206] S. Lee, A. Liu, Z. J. Wang, and M. J. McKeown. Abnormal Phase Cou-pling in Parkinsons Disease and Normalization Effects of SubthresholdVestibular Stimulation. Frontiers in Human Neuroscience, 13:118, 42019.[207] W. H. Lee and S. Frangou. Linking functional connectivity anddynamic properties of resting-state networks. Scientific Reports,7(1):16610, 12 2017.[208] W. Legon, T. F. Sato, A. Opitz, J. Mueller, A. Barbour, A. Williams,and W. J. Tyler. Transcranial focused ultrasound modulates the activ-ity of primary somatosensory cortex in humans. Nature Neuroscience,17(2):322–329, 2 2014.[209] B. Lenggenhager, C. Lopez, and O. Blanke. Influence of galvanicvestibular stimulation on egocentric and object-based mental trans-formations. Experimental Brain Research, 184(2):211–221, 11 2007.[210] D. K. Leventhal, G. J. Gage, R. Schmidt, J. R. Pettibone, A. C. Case,and J. D. Berke. Basal Ganglia Beta Oscillations Accompany CueUtilization. Neuron, 73(3):523–536, 2 2012.[211] R. Levy, P. Ashby, W. D. Hutchison, A. E. Lang, A. M. Lozano, andJ. O. Dostrovsky. Dependence of subthalamic nucleus oscillations onmovement and dopamine in Parkinson’s disease. Brain : a journal ofneurology, 125(Pt 6):1196–209, 6 2002.[212] R. Levy, L. N. Hazrati, M. T. Herrero, M. Vila, O. K. Hassani,M. Mouroux, M. Ruberg, H. Asensi, Y. Agid, J. Fe´ger, J. A. Obeso,A. Parent, and E. C. Hirsch. Re-evaluation of the functional anatomyof the basal ganglia in normal and Parkinsonian states. Neuroscience,76(2):335–43, 1 1997.151Bibliography[213] R. Levy, W. D. Hutchison, A. M. Lozano, and J. O. Dostrovsky. High-frequency Synchronization of Neuronal Activity in the SubthalamicNucleus of Parkinsonian Patients with Limb Tremor. Journal of Neu-roscience, 20(20), 2000.[214] R. Levy, W. D. Hutchison, A. M. Lozano, and J. O. Dostrovsky. Syn-chronized neuronal discharge in the basal ganglia of parkinsonian pa-tients is limited to oscillatory activity. The Journal of neuroscience :the official journal of the Society for Neuroscience, 22(7):2855–61, 42002.[215] W. Li, J. Liu, F. Skidmore, Y. Liu, J. Tian, and K. Li. White MatterMicrostructure Changes in the Thalamus in Parkinson Disease withDepression: A Diffusion Tensor MR Imaging Study. American Journalof Neuroradiology, 31(10):1861–1866, 11 2010.[216] K. Lieb, S. Brucker, M. Bach, T. Els, C. H. Lu¨cking, and M. W.Greenlee. Impairment in preattentive visual processing in patientswith Parkinson’s disease. Brain, 122(2):303–313, 2 1999.[217] D. Liebetanz, M. A. Nitsche, F. Tergau, and W. Paulus. Pharmaco-logical approach to the mechanisms of transcranial DC-stimulation-induced after-effects of human motor cortex excitability. Brain : ajournal of neurology, 125(Pt 10):2238–47, 10 2002.[218] S. Lippe, N. Kovacevic, and A. R. McIntosh. Differential maturationof brain signal complexity in the human auditory and visual system.Frontiers in Human Neuroscience, 3:48, 2009.[219] V. Litvak, A. Eusebio, A. Jha, R. Oostenveld, G. Barnes, T. Foltynie,P. Limousin, L. Zrinzo, M. I. Hariz, K. Friston, and P. Brown.Movement-related changes in local and long-range synchronization inParkinson’s disease revealed by simultaneous magnetoencephalogra-phy and intracranial recordings. The Journal of neuroscience : theofficial journal of the Society for Neuroscience, 32(31):10541–53, 82012.[220] E. Lobel, J. F. Kleine, D. L. Bihan, A. Leroy-Willig, and A. Berthoz.Functional MRI of galvanic vestibular stimulation. Journal of neuro-physiology, 80(5):2699–709, 11 1998.[221] C. Lopez and O. Blanke. The thalamocortical vestibular system in an-imals and humans. Brain Research Reviews, 67(1-2):119–146, 6 2011.152Bibliography[222] C. Lopez, O. Blanke, and F. Mast. The human vestibular cortexrevealed by coordinate-based activation likelihood estimation meta-analysis. Neuroscience, 212:159–179, 6 2012.[223] S. Luan, I. Williams, K. Nikolic, and T. G. Constandinou. Neuromodu-lation: present and emerging methods. Frontiers in Neuroengineering,7:27, 7 2014.[224] C. D. B. Luft, E. Pereda, M. J. Banissy, and J. Bhattacharya. Bestof both worlds: promise of combining brain stimulation and brainconnectome. Frontiers in Systems Neuroscience, 8:132, 7 2014.[225] K. E. Lyons, S. B. Wilkinson, J. Overman, and R. Pahwa. Surgicaland hardware complications of subthalamic stimulation: a series of160 procedures. Neurology, 63(4):612–6, 8 2004.[226] A. D. Macleod, K. S. Taylor, and C. E. Counsell. Mortality in Parkin-son’s disease: A systematic review and meta-analysis. Movement Dis-orders, 29(13):1615–1622, 11 2014.[227] G. Magnani, M. Cursi, L. Leocani, M. A. Volonte´, and G. Comi. Acuteeffects of L-dopa on event-related desynchronization in Parkinson’sdisease. Neurological Sciences, 23(3):91–97, 9 2002.[228] Q. Mai. A review of discriminant analysis in high dimensions. WileyInterdisciplinary Reviews: Computational Statistics, 5(3):190–197, 52013.[229] E. Mak, L. Su, G. B. Williams, M. J. Firbank, R. A. Lawson, A. J.Yarnall, G. W. Duncan, A. M. Owen, T. K. Khoo, D. J. Brooks,J. B. Rowe, R. A. Barker, D. J. Burn, and J. T. OBrien. Baselineand longitudinal grey matter changes in newly diagnosed Parkinsonsdisease: ICICLE-PD study. Brain, 138(10):2974–2986, 10 2015.[230] R. Manenti, M. Brambilla, S. Rosini, I. Orizio, C. Ferrari, B. Borroni,and M. Cotelli. Time up and go task performance improves after tran-scranial direct current stimulation in patient affected by Parkinson’sdisease. Neuroscience Letters, 580:74–77, 9 2014.[231] E. Manjarrez, G. Rojas-Piloni, I. Me´ndez, and A. Flores. Stochasticresonance within the somatosensory system: effects of noise on evokedfield potentials elicited by tactile stimuli. The Journal of neuroscience: the official journal of the Society for Neuroscience, 23(6):1997–2001,3 2003.153Bibliography[232] F. Mars, K. Popov, and J. L. Vercher. Supramodal effects of gal-vanic vestibular stimulation on the subjective vertical. Neuroreport,12(13):2991–4, 9 2001.[233] J. F. Marsden, P. Limousin-Dowsey, P. Ashby, P. Pollak, andP. Brown. Subthalamic nucleus, sensorimotor cortex and muscle in-terrelationships in Parkinson’s disease. Brain : a journal of neurology,124(Pt 2):378–88, 2 2001.[234] K. Matsunami and B. Cohen. Afferent modulation of unit activity inglobus pallidus and caudate nucleus: changes induced by vestibularnucleus and pyramidal tract stimulation. Brain Research, 91(1):140–146, 6 1975.[235] V. S. Mattay, J. H. Callicott, A. Bertolino, A. K. Santha, J. D.Van Horn, K. A. Tallent, J. A. Frank, and D. R. Weinberger. Hemi-spheric control of motor function: a whole brain echo planar fMRIstudy. Psychiatry research, 83(1):7–22, 7 1998.[236] P. Mazzoni, B. Shabbott, and J. C. Corte´s. Motor control abnor-malities in Parkinson’s disease. Cold Spring Harbor perspectives inmedicine, 2(6):a009282, 6 2012.[237] M. D. McDonnell and D. Abbott. What is stochastic resonance? Def-initions, misconceptions, debates, and its relevance to biology. PLoScomputational biology, 5(5):e1000348, 5 2009.[238] M. D. McDonnell and L. M. Ward. The benefits of noise in neural sys-tems: bridging theory and experiment. Nature Reviews Neuroscience,12(7):415–426, 6 2011.[239] E. G. McGeer, W. A. Staines, and P. L. McGeer. Neurotransmittersin the basal ganglia. The Canadian journal of neurological sciences.Le journal canadien des sciences neurologiques, 11(1 Suppl):89–99, 21984.[240] P. D. McGeoch and V. S. Ramachandran. Vestibular stimulation canrelieve central pain of spinal origin. Spinal Cord, 46(11):756–757, 112008.[241] P. D. McGeoch, L. E. Williams, R. R. Lee, and V. S. Ramachan-dran. Behavioural evidence for vestibular stimulation as a treatmentfor central post-stroke pain. Journal of Neurology, Neurosurgery &Psychiatry, 79(11):1298–1301, 6 2008.154Bibliography[242] J. G. McHaffie, T. R. Stanford, B. E. Stein, V. Coizet, and P. Red-grave. Subcortical loops through the basal ganglia. Trends in Neuro-sciences, 28(8):401–407, 8 2005.[243] A. R. McIntosh, N. Kovacevic, and R. J. Itier. Increased brain signalvariability accompanies lower behavioral variability in development.PLoS computational biology, 4(7):e1000106, 7 2008.[244] C. C. McIntyre, M. Savasta, L. Kerkerian-Le Goff, and J. L. Vitek. Un-covering the mechanism(s) of action of deep brain stimulation: activa-tion, inhibition, or both. Clinical Neurophysiology, 115(6):1239–1248,6 2004.[245] M. J. McKeown, A. Uthama, R. Abugharbieh, S. Palmer, M. Lewis,and X. Huang. Shape (but not volume) changes in the thalami inParkinson disease. BMC Neurology, 8(1):8, 12 2008.[246] I. Mendez-Balbuena, E. Manjarrez, J. Schulte-Monting, F. Huethe,J. A. Tapia, M.-C. Hepp-Reymond, and R. Kristeva. Improved Sen-sorimotor Performance via Stochastic Resonance. Journal of Neuro-science, 32(36):12612–12618, 9 2012.[247] H. Meng, P. J. May, J. D. Dickman, and D. E. Angelaki. VestibularSignals in Primate Thalamus: Properties and Origins. Journal ofNeuroscience, 27(50):13590–13602, 12 2007.[248] R. A. Menke, J. Scholz, K. L. Miller, S. Deoni, S. Jbabdi, P. M.Matthews, and M. Zarei. MRI characteristics of the substantia nigrain Parkinson’s disease: A combined quantitative T1 and DTI study.NeuroImage, 47(2):435–441, 8 2009.[249] C. Miniussi, D. Brignani, and M. C. Pellicciari. Combining Tran-scranial Electrical Stimulation With Electroencephalography. ClinicalEEG and Neuroscience, 43(3):184–191, 7 2012.[250] C. Miniussi, J. A. Harris, and M. Ruzzoli. Modelling non-invasive brainstimulation in cognitive neuroscience. Neuroscience & BiobehavioralReviews, 37(8):1702–1712, 9 2013.[251] A. S. Mitchell and S. Chakraborty. What does the mediodorsal tha-lamus do? Frontiers in systems neuroscience, 7:37, 2013.155Bibliography[252] D. J. Mitchell, N. McNaughton, D. Flanagan, and I. J. Kirk. Frontal-midline theta from the perspective of hippocampal theta. Progress inNeurobiology, 86(3):156–185, 11 2008.[253] Y. Miyagi. Thalamic Stimulation for Parkinsons Disease: ClinicalStudies on DBS. In Deep Brain Stimulation for Neurological Disorders,pages 103–120. Springer International Publishing, Cham, 2015.[254] M. Moazami-Goudarzi, J. Sarnthein, L. Michels, R. Moukhtieva, andD. Jeanmonod. Enhanced frontal low and high frequency power andsynchronization in the resting EEG of parkinsonian patients. Neu-roImage, 41(3):985–97, 7 2008.[255] K. Monte-Silva, M.-F. Kuo, S. Hessenthaler, S. Fresnoza, D. Liebetanz,W. Paulus, and M. A. Nitsche. Induction of Late LTP-Like Plasticityin the Human Motor Cortex by Repeated Non-Invasive Brain Stimu-lation. Brain Stimulation, 6(3):424–432, 5 2013.[256] F. Mormann, K. Lehnertz, P. David, and C. E. Elger. Mean phasecoherence as a measure for phase synchronization and its applicationto the EEG of epilepsy patients. Physica D: Nonlinear Phenomena,144(3-4):358–369, 10 2000.[257] F. Moss, L. M. Ward, and W. G. Sannita. Stochastic resonance andsensory information processing: a tutorial and review of application.Clinical neurophysiology : official journal of the International Federa-tion of Clinical Neurophysiology, 115(2):267–81, 2 2004.[258] D. Muslimovic, B. Schmand, J. D. Speelman, and R. J. De Haan.Course of cognitive decline in Parkinson’s disease: A meta-analysis.Journal of the International Neuropsychological Society, 13(06):920–32, 11 2007.[259] A. B. Nelson, C. Moisello, J. Lin, P. Panday, S. Ricci, A. Canessa,A. Di Rocco, A. Quartarone, G. Frazzitta, I. U. Isaias, G. Tononi,C. Cirelli, and M. F. Ghilardi. Beta Oscillatory Changes and Retentionof Motor Skills during Practice in Healthy Subjects and in Patientswith Parkinson’s Disease. Frontiers in Human Neuroscience, 11:104,3 2017.[260] M. Y. Neufeld, R. Inzelberg, and A. D. Korczyn. EEG in demented andnon-demented parkinsonian patients. Acta neurologica Scandinavica,78(1):1–5, 7 1988.156Bibliography[261] T. Neuling, S. Rach, and C. S. Herrmann. Orchestrating neuronal net-works: sustained after-effects of transcranial alternating current stim-ulation depend upon brain states. Frontiers in Human Neuroscience,7:161, 2013.[262] S. M. Nicola, D. J. Surmeier, and R. C. Malenka. Dopaminergic Mod-ulation of Neuronal Excitability in the Striatum and Nucleus Accum-bens. Annual Review of Neuroscience, 23(1):185–215, 3 2000.[263] F. Nieuwhof, B. R. Bloem, M. F. Reelick, E. Aarts, I. Maidan,A. Mirelman, J. M. Hausdorff, I. Toni, and R. C. Helmich. Impaireddual tasking in Parkinsons disease is associated with reduced focusingof cortico-striatal activity. Brain, 140(5):1384–1398, 5 2017.[264] M. A. Nitsche and W. Paulus. Excitability changes induced in thehuman motor cortex by weak transcranial direct current stimulation.The Journal of physiology, 527 Pt 3:633–9, 9 2000.[265] Y. Niv, N. D. Daw, D. Joel, and P. Dayan. Tonic dopamine: oppor-tunity costs and the control of response vigor. Psychopharmacology,191(3):507–520, 3 2007.[266] G. Nolte, O. Bai, L. Wheaton, Z. Mari, S. Vorbach, and M. Hallett.Identifying true brain interaction from EEG data using the imaginarypart of coherency. Clinical neurophysiology : official journal of theInternational Federation of Clinical Neurophysiology, 115(10):2292–307, 10 2004.[267] A. Notbohm, J. Kurths, and C. S. Herrmann. Modification of BrainOscillations via Rhythmic Light Stimulation Provides Evidence forEntrainment but Not for Superposition of Event-Related Responses.Frontiers in human neuroscience, 10:10, 2016.[268] N. Noury, J. F. Hipp, and M. Siegel. Physiological processes non-linearly affect electrophysiological recordings during transcranial elec-tric stimulation. NeuroImage, 140:99–109, 10 2016.[269] N. Noury and M. Siegel. Phase properties of transcranial electricalstimulation artifacts in electrophysiological recordings. NeuroImage,158:406–416, 9 2017.[270] M. Nowak, C. Zich, and C. J. Stagg. Motor Cortical Gamma Oscilla-tions: What Have We Learnt and Where Are We Headed? Currentbehavioral neuroscience reports, 5(2):136–142, 2018.157Bibliography[271] M. Oh, J. S. Kim, J. Y. Kim, K.-H. Shin, S. H. Park, H. O. Kim, D. H.Moon, S. J. Oh, S. J. Chung, and C. S. Lee. Subregional Patterns ofPreferential Striatal Dopamine Transporter Loss Differ in ParkinsonDisease, Progressive Supranuclear Palsy, and Multiple-System Atro-phy. Journal of Nuclear Medicine, 53(3):399–406, 3 2012.[272] Y. Okada, Y. Kita, J. Nakamura, H. Kataoka, T. Kiriyama, S. Ueno,M. Hiyamizu, S. Morioka, and K. Shomoto. Galvanic vestibular stim-ulation may improve anterior bending posture in Parkinson’s disease.Neuroreport, 26(7):405–10, 5 2015.[273] M. S. Okun. Deep-Brain Stimulation for Parkinson’s Disease. NewEngland Journal of Medicine, 367(16):1529–1538, 10 2012.[274] M. S. Okun, B. V. Gallo, G. Mandybur, J. Jagid, K. D. Foote, F. J.Revilla, R. Alterman, J. Jankovic, R. Simpson, F. Junn, L. Verhagen,J. E. Arle, B. Ford, R. R. Goodman, R. M. Stewart, S. Horn, G. H.Baltuch, B. H. Kopell, F. Marshall, D. Peichel, R. Pahwa, K. E. Lyons,A. I. Tro¨ster, J. L. Vitek, M. Tagliati, and SJM DBS Study Group.Subthalamic deep brain stimulation with a constant-current device inParkinson’s disease: an open-label randomised controlled trial. TheLancet Neurology, 11(2):140–149, 2 2012.[275] A. Oswal, P. Brown, and V. Litvak. Synchronized neural oscillationsand the pathophysiology of Parkinson’s disease. Current opinion inneurology, 26(6):662–70, 12 2013.[276] M. L. Otten, C. B. Mikell, B. E. Youngerman, C. Liston, M. B. Sisti,J. N. Bruce, S. A. Small, and G. M. McKhann. Motor deficits correlatewith resting state motor network connectivity in patients with braintumours. Brain, 135(4):1017–1026, 4 2012.[277] S. Pal, S. M. Rosengren, and J. G. Colebatch. Stochastic galvanicvestibular stimulation produces a small reduction in sway in Parkin-son’s disease. Journal of vestibular research : equilibrium & orienta-tion, 19(3-4):137–42, 2009.[278] W. Pan, R. Soma, S. Kwak, and Y. Yamamoto. Improvement of motorfunctions by noisy vestibular stimulation in central neurodegenerativedisorders. Journal of neurology, 255(11):1657–61, 11 2008.[279] J. Parkinson. An Essay on the Shaking Palsy. The Journal of Neu-ropsychiatry and Clinical Neurosciences, 14(2):223–236, 5 2002.158Bibliography[280] S. Parnaudeau, S. S. Bolkan, and C. Kellendonk. The MediodorsalThalamus: An Essential Partner of the Prefrontal Cortex for Cogni-tion. Biological Psychiatry, 83(8):648–656, 4 2018.[281] S. Parnaudeau, P.-K. O’Neill, S. S. Bolkan, R. D. Ward, A. I. Abbas,B. L. Roth, P. D. Balsam, J. A. Gordon, and C. Kellendonk. Inhi-bition of mediodorsal thalamus disrupts thalamofrontal connectivityand cognition. Neuron, 77(6):1151–62, 3 2013.[282] T. D. Parsons, S. A. Rogers, A. J. Braaten, S. P. Woods, and A. I.Tro¨ster. Cognitive sequelae of subthalamic nucleus deep brain stimu-lation in Parkinson’s disease: a meta-analysis. The Lancet Neurology,5(7):578–588, 7 2006.[283] M. A. Pastor, B. L. Day, and C. D. Marsden. Vestibular induced pos-tural responses in Parkinson’s disease. Brain : a journal of neurology,116 ( Pt 5:1177–90, 10 1993.[284] S. J. Pelletier and F. Cicchetti. Cellular and molecular mechanisms ofaction of transcranial direct current stimulation: evidence from in vitroand in vivo models. The international journal of neuropsychopharma-cology, 18(2), 10 2014.[285] M. Penttila¨, J. V. Partanen, H. Soininen, and P. J. Riekkinen. Quan-titative analysis of occipital EEG in different stages of Alzheimer’sdisease. Electroencephalography and clinical neurophysiology, 60(1):1–6, 1 1985.[286] G. Percheron, C. Franc¸ois, B. Talbi, J. Yelnik, and G. Fe´nelon. Theprimate motor thalamus. Brain research. Brain research reviews,22(2):93–181, 8 1996.[287] J. B. Pereira, C. Junque´, D. Bartre´s-Faz, M. J. Mart´ı, R. Sala-Llonch,Y. Compta, C. Falco´n, P. Vendrell, . Pascual-Leone, J. Valls-Sole´, andE. Tolosa. Modulation of verbal fluency networks by transcranial directcurrent stimulation (tDCS) in Parkinsons disease. Brain Stimulation,6(1):16–24, 1 2013.[288] S. Perez-Lloret and F. J. Barrantes. Deficits in cholinergic neuro-transmission and their clinical correlates in Parkinson’s disease. NPJParkinson’s disease, 2:16001, 2016.159Bibliography[289] G. Pfurtscheller, C. Brunner, A. Schlo¨gl, and F. Lopes da Silva. Murhythm (de)synchronization and EEG single-trial classification of dif-ferent motor imagery tasks. NeuroImage, 31(1):153–159, 5 2006.[290] G. Pfurtscheller, B. Graimann, J. E. Huggins, S. P. Levine, and L. A.Schuh. Spatiotemporal patterns of beta desynchronization and gammasynchronization in corticographic data during self-paced movement.Clinical neurophysiology : official journal of the International Federa-tion of Clinical Neurophysiology, 114(7):1226–36, 7 2003.[291] G. Pfurtscheller and F. H. Lopes da Silva. Event-related EEG/MEGsynchronization and desynchronization: basic principles. Clinical neu-rophysiology : official journal of the International Federation of Clin-ical Neurophysiology, 110(11):1842–57, 11 1999.[292] G. Pfurtscheller, C. Neuper, and J. Kalcher. 40-Hz oscillations dur-ing motor behavior in man. Neuroscience letters, 164(1-2):179–82, 121993.[293] A. Pikovsky, M. Rosenblum, and J. Kurths. Synchronization: a uni-versal concept in nonlinear sciences. Cambridge University Press,2003.[294] R. Pintelon and J. Schoukens. System identification: a frequency do-main approach. John Wiley & Sons, Inc., Hoboken, NJ, USA, 3 2012.[295] B. Pinter, A. Diem-Zangerl, G. K. Wenning, C. Scherfler,W. Oberaigner, K. Seppi, and W. Poewe. Mortality in Parkinson’sdisease: A 38-year follow-up study. Movement Disorders, 30(2):266–269, 2 2015.[296] C. Pirulli, A. Fertonani, and C. Miniussi. The Role of Timing in theInduction of Neuromodulation in Perceptual Learning by TranscranialElectric Stimulation. Brain Stimulation, 6(4):683–689, 7 2013.[297] W. Poewe, K. Seppi, C. M. Tanner, G. M. Halliday, P. Brundin,J. Volkmann, A.-E. Schrag, and A. E. Lang. Parkinson disease. NatureReviews Disease Primers, 3:17013, 3 2017.[298] A. Pogosyan, L. D. Gaynor, A. Eusebio, and P. Brown. Boosting Cor-tical Activity at Beta-Band Frequencies Slows Movement in Humans.Current Biology, 19(19):1637–1641, 10 2009.160Bibliography[299] R. Polan´ıa, M. A. Nitsche, and W. Paulus. Modulating functionalconnectivity patterns and topological functional organization of thehuman brain with transcranial direct current stimulation. HumanBrain Mapping, 32(8):1236–1249, 8 2011.[300] R. Polan´ıa, M. A. Nitsche, and C. C. Ruff. Studying and modifyingbrain function with non-invasive brain stimulation. Nature Neuro-science, 21(2):174–187, 2 2018.[301] B. Pollok, D. Kamp, M. Butz, L. Wojtecki, L. Timmermann,M. Su¨dmeyer, V. Krause, and A. Schnitzler. Increased SMA-M1 co-herence in Parkinson’s disease - Pathophysiology or compensation?Experimental neurology, 247:178–81, 9 2013.[302] G. Pontone, J. R. Williams, S. S. Bassett, and L. Marsh. Clinical fea-tures associated with impulse control disorders in Parkinson disease.Neurology, 67(7):1258–1261, 10 2006.[303] B. D. Prakash, Y.-Y. Sitoh, L. C. Tan, and W. L. Au. Asymmetri-cal diffusion tensor imaging indices of the rostral substantia nigra inParkinson’s disease. Parkinsonism & Related Disorders, 18(9):1029–1033, 11 2012.[304] M. J. Price, R. G. Feldman, D. Adelberg, and H. Kayne. Abnor-malities in color vision and contrast sensitivity in Parkinson’s disease.Neurology, 42(4):887–90, 4 1992.[305] T. Pringsheim, N. Jette, A. Frolkis, and T. D. Steeves. The preva-lence of Parkinson’s disease: A systematic review and meta-analysis.Movement Disorders, 29(13):1583–1590, 11 2014.[306] A. Priori, A. Berardelli, S. Rona, N. Accornero, and M. Manfredi. Po-larization of the human motor cortex through the scalp. Neuroreport,9(10):2257–60, 7 1998.[307] A. Priori, G. Foffani, A. Pesenti, F. Tamma, A. Bianchi, M. Pelle-grini, M. Locatelli, K. Moxon, and R. Villani. Rhythm-specific phar-macological modulation of subthalamic activity in Parkinson’s disease.Experimental Neurology, 189(2):369–379, 10 2004.[308] A. Priori, G. Foffani, L. Rossi, and S. Marceglia. Adaptive deep brainstimulation (aDBS) controlled by local field potential oscillations. Ex-perimental Neurology, 245:77–86, 7 2013.161Bibliography[309] M. V. Puig, A. Watakabe, M. Ushimaru, T. Yamamori, andY. Kawaguchi. Serotonin Modulates Fast-Spiking Interneuron andSynchronous Activity in the Rat Prefrontal Cortex through 5-HT1Aand 5-HT2A Receptors. Journal of Neuroscience, 30(6):2211–2222, 22010.[310] J. D. Putzke, R. E. Wharen, Z. K. Wszolek, M. F. Turk, A. J. Stron-gosky, and R. J. Uitti. Thalamic deep brain stimulation for tremor-predominant Parkinson’s disease. Parkinsonism & related disorders,10(2):81–8, 12 2003.[311] N. Pyatigorskaya, C. Gallea, D. Garcia-Lorenzo, M. Vidailhet, andS. Lehericy. A review of the use of magnetic resonance imaging inParkinson’s disease. Therapeutic advances in neurological disorders,7(4):206–20, 7 2014.[312] A. Rajagopalan, K. V. Jinu, K. S. Sailesh, S. Mishra, U. K. Reddy,and J. K. Mukkadan. Understanding the links between vestibularand limbic systems regulating emotions. Journal of natural science,biology, and medicine, 8(1):11–15, 2017.[313] A. H. Rajput, H. H. Sitte, A. Rajput, M. E. Fenton, C. Pifl, andO. Hornykiewicz. Globus pallidus dopamine and Parkinson motor sub-types: Clinical and brain biochemical correlation. Neurology, 70(Issue16, Part 2):1403–1410, 4 2008.[314] E. A. Rancz, J. Moya, F. Drawitsch, A. M. Brichta, S. Canals, andT. W. Margrie. Widespread vestibular activation of the rodent cortex.The Journal of neuroscience : the official journal of the Society forNeuroscience, 35(15):5926–34, 4 2015.[315] J. S. Richman and J. R. Moorman. Physiological time-series analysisusing approximate entropy and sample entropy. American Journal ofPhysiology-Heart and Circulatory Physiology, 278(6):H2039–H2049, 62000.[316] L. Ris, M. Hachemaoui, N. Vibert, E. Godaux, P. P. Vidal, and L. E.Moore. Resonance of Spike Discharge Modulation in Neurons of theGuinea Pig Medial Vestibular Nucleus. Journal of Neurophysiology,86(2):703–716, 8 2001.[317] M. C. Rodriguez-Oroz, M. Jahanshahi, P. Krack, I. Litvan, R. Ma-cias, E. Bezard, and J. A. Obeso. Initial clinical manifestations of162BibliographyParkinson’s disease: features and pathophysiological mechanisms. TheLancet Neurology, 8(12):1128–1139, 12 2009.[318] M. C. Rodriguez-Oroz, J. A. Obeso, A. E. Lang, J.-L. Houeto,P. Pollak, S. Rehncrona, J. Kulisevsky, A. Albanese, J. Volkmann,M. I. Hariz, N. P. Quinn, J. D. Speelman, J. Guridi, I. Zamarbide,A. Gironell, J. Molet, B. Pascual-Sedano, B. Pidoux, A. M. Bonnet,Y. Agid, J. Xie, A.-L. Benabid, A. M. Lozano, J. Saint-Cyr, L. Romito,M. F. Contarino, M. Scerrati, V. Fraix, and N. Van Blercom. Bilat-eral deep brain stimulation in Parkinson’s disease: a multicentre studywith 4 years follow-up. Brain, 128(10):2240–2249, 10 2005.[319] S. Rossi, M. Hallett, P. M. Rossini, A. Pascual-Leone, and Safety ofTMS Consensus Group. Safety, ethical considerations, and applicationguidelines for the use of transcranial magnetic stimulation in clinicalpractice and research. Clinical Neurophysiology, 120(12):2008–2039,12 2009.[320] P. Rossini, L. Rosinni, and F. Ferreri. Brain-Behavior Relations:Transcranial Magnetic Stimulation: A Review. IEEE Engineering inMedicine and Biology Magazine, 29(1):84–96, 1 2010.[321] A. Roy, B. Baxter, and B. He. High-definition transcranial directcurrent stimulation induces both acute and persistent changes inbroadband cortical synchronization: a simultaneous tDCS-EEG study.IEEE transactions on bio-medical engineering, 61(7):1967–78, 7 2014.[322] K. S. Rufener, P. Ruhnau, H.-J. Heinze, and T. Zaehle. TranscranialRandom Noise Stimulation (tRNS) Shapes the Processing of RapidlyChanging Auditory Information. Frontiers in cellular neuroscience,11:162, 2017.[323] G. Ruffini, M. D. Fox, O. Ripolles, P. C. Miranda, and A. Pascual-Leone. Optimization of multifocal transcranial current stimulation forweighted cortical pattern targeting from realistic modeling of electricfields. NeuroImage, 89:216–225, 4 2014.[324] Y. U. Ryu and J. J. Buchanan. Accuracy, Stability, and CorrectiveBehavior in a Visuomotor Tracking Task: A Preliminary Study. PLoSONE, 7(6):e38537, 6 2012.163Bibliography[325] R. J. Sadleir, T. D. Vannorsdall, D. J. Schretlen, and B. Gordon. Tran-scranial direct current stimulation (tDCS) in a realistic head model.NeuroImage, 51(4):1310–1318, 7 2010.[326] S. E. Safo and Q. Long. Sparse linear discriminant analysis in struc-tured covariates space. Statistical Analysis and Data Mining: TheASA Data Science Journal, 4 2018.[327] J. D. Salamone, M. Correa, A. M. Farrar, E. J. Nunes, and M. Pardo.Dopamine, Behavioral Economics, and Effort. Frontiers in BehavioralNeuroscience, 3:13, 2009.[328] S. Salenius, R. Salmelin, C. Neuper, G. Pfurtscheller, and R. Hari.Human cortical 40 Hz rhythm is closely related to EMG rhythmicity.Neuroscience letters, 213(2):75–8, 8 1996.[329] Y. Salimpour, Z. K. Mari, and R. Shadmehr. Altering Effort Costsin Parkinson’s Disease with Noninvasive Cortical Stimulation. TheJournal of neuroscience : the official journal of the Society for Neu-roscience, 35(35):12287–302, 9 2015.[330] G. Samoudi, M. Jiveg˚ard, A. P. Mulavara, and F. Bergquist. Effects ofStochastic Vestibular Galvanic Stimulation and LDOPA on Balanceand Motor Symptoms in Patients With Parkinson’s Disease. BrainStimulation, 8(3):474–480, 5 2015.[331] G. Samoudi, H. Nissbrandt, M. B. Dutia, and F. Bergquist. NoisyGalvanic Vestibular Stimulation Promotes GABA Release in the Sub-stantia Nigra and Improves Locomotion in Hemiparkinsonian Rats.PLoS ONE, 7(1):e29308, 1 2012.[332] J. Sarnthein and D. Jeanmonod. High thalamocortical theta coherencein patients with Parkinson’s disease. The Journal of neuroscience : theofficial journal of the Society for Neuroscience, 27(1):124–31, 1 2007.[333] P. Sauseng and W. Klimesch. What does phase information of oscil-latory brain activity tell us about cognitive processes? Neuroscience& Biobehavioral Reviews, 32(5):1001–1013, 7 2008.[334] R. Savica, B. R. Grossardt, J. H. Bower, J. E. Ahlskog, and W. A.Rocca. Incidence and Pathology of Synucleinopathies and TauopathiesRelated to Parkinsonism. JAMA Neurology, 70(7):859, 7 2013.164Bibliography[335] T. A. Scandalis, A. Bosak, J. C. Berliner, L. L. Helman, and M. R.Wells. Resistance training and gait function in patients with Parkin-son’s disease. American journal of physical medicine & rehabilitation,80(1):38–43, 1 2001.[336] C. Scherfler, K. Seppi, K. J. Mair, E. Donnemiller, I. Virgolini, G. K.Wenning, and W. Poewe. Left hemispheric predominance of nigrostri-atal dysfunction in Parkinsons disease. Brain, 135(11):3348–3354, 112012.[337] W. Schlee, M. Schecklmann, A. Lehner, P. M. Kreuzer, V. Vielsmeier,T. B. Poeppl, and B. Langguth. Reduced variability of auditory alphaactivity in chronic tinnitus. Neural plasticity, 2014:436146, 2014.[338] J. E. Schlerf, J. M. Galea, D. Spampinato, and P. A. Celnik. LateralityDifferences in CerebellarMotor Cortex Connectivity. Cerebral Cortex,25(7):1827–1834, 7 2015.[339] L. Schmidt, K. S. Utz, L. Depper, M. Adams, A.-K. Schaadt, S. Rein-hart, and G. Kerkhoff. Now You Feel both: Galvanic Vestibular Stimu-lation Induces Lasting Improvements in the Rehabilitation of ChronicTactile Extinction. Frontiers in Human Neuroscience, 7:90, 2013.[340] E. Schneider, S. Glasauer, and M. Dieterich. Comparison of Hu-man Ocular Torsion Patterns During Natural and Galvanic VestibularStimulation. Journal of Neurophysiology, 87(4):2064–2073, 4 2002.[341] J. Schoukens, P. Guillaume, and R. Pintelon. Perturbation signals forsystem identification. Prentice Hall, 1993.[342] A. Schouten, E. de Vlugt, and F. van der Helm. Design of Pertur-bation Signals for the Estimation of Proprioceptive Reflexes. IEEETransactions on Biomedical Engineering, 55(5):1612–1619, 5 2008.[343] M. Schreckenberger, C. Lange-Asschenfeldt, C. Lange-Asschenfeld,M. Lochmann, K. Mann, T. Siessmeier, H.-G. Buchholz, P. Barten-stein, G. Gru¨nder, and G. Gru¨nder. The thalamus as the generatorand modulator of EEG alpha rhythm: a combined PET/EEG studywith lorazepam challenge in humans. NeuroImage, 22(2):637–44, 62004.[344] T. M. Seibert, E. A. Murphy, E. J. Kaestner, and J. B. Brewer. Inter-regional Correlations in Parkinson Disease and Parkinson-related De-165Bibliographymentia with Resting Functional MR Imaging. Radiology, 263(1):226–234, 4 2012.[345] K. Sethi. Levodopa unresponsive symptoms in Parkinson disease.Movement Disorders, 23(S3):S521–S533, 1 2008.[346] A. Se´verac Cauquil, M. Faldon, K. Popov, B. L. Day, and A. M. Bron-stein. Short-latency eye movements evoked by near-threshold galvanicvestibular stimulation. Experimental brain research, 148(3):414–8, 22003.[347] M. Sharman, R. Valabregue, V. Perlbarg, L. Marrakchi-Kacem,M. Vidailhet, H. Benali, A. Brice, and S. Lehe´ricy. Parkinson’s diseasepatients show reduced cortical-subcortical sensorimotor connectivity.Movement Disorders, 28(4):447–454, 4 2013.[348] H. A. Shill, S. Obradov, Y. Katsnelson, and R. Pizinger. A ran-domized, double-blind trial of transcranial electrostimulation in earlyParkinson’s disease. Movement Disorders, 26(8):1477–1480, 7 2011.[349] S. Shin, J. E. Lee, J. Y. Hong, M.-K. Sunwoo, Y. H. Sohn, and P. H.Lee. Neuroanatomical substrates of visual hallucinations in patientswith non-demented Parkinson’s disease. Journal of neurology, neuro-surgery, and psychiatry, 83(12):1155–61, 12 2012.[350] P. Silberstein, A. A. Ku¨hn, A. Kupsch, T. Trottenberg, J. K. Krauss,J. C. Wo¨hrle, P. Mazzone, A. Insola, V. Di Lazzaro, A. Oliviero,T. Aziz, and P. Brown. Patterning of globus pallidus local fieldpotentials differs between Parkinson’s disease and dystonia. Brain,126(12):2597–2608, 9 2003.[351] P. Silberstein, A. Pogosyan, A. A. Ku¨hn, G. Hotton, S. Tisch, A. Kup-sch, P. Dowsey-Limousin, M. I. Hariz, and P. Brown. Cortico-corticalcoupling in Parkinson’s disease and its modulation by therapy. Brain,128(Pt 6):1277–91, 6 2005.[352] R. F. Silva, S. M. Plis, J. Sui, M. S. Pattichis, T. Adali, and V. D. Cal-houn. Blind Source Separation for Unimodal and Multimodal BrainNetworks: A Unifying Framework for Subspace Modeling. IEEE Jour-nal of Selected Topics in Signal Processing, 10(7):1134–1149, 10 2016.[353] G. Simon and J. Schoukens. Robust broadband periodic excitationdesign. IEEE Transactions on Instrumentation and Measurement,49(2):270–274, 4 2000.166Bibliography[354] H. A. Slagter, A. Mazaheri, L. C. Reteig, R. Smolders, M. Figee,M. Mantione, P. R. Schuurman, and D. Denys. Contributionsof the Ventral Striatum to Conscious Perception: An IntracranialEEG Study of the Attentional Blink. The Journal of Neuroscience,37(5):1081–1089, 2 2017.[355] R. Smolders, A. Mazaheri, G. van Wingen, M. Figee, P. P. de Kon-ing, and D. Denys. Deep Brain Stimulation Targeted at the NucleusAccumbens Decreases the Potential for Pathologic Network Commu-nication. Biological Psychiatry, 74(10):e27–e28, 11 2013.[356] V. S. Sohal. Insights into Cortical Oscillations Arising from Optoge-netic Studies. Biological Psychiatry, 71(12):1039–1045, 6 2012.[357] R. Soma, S. Kwak, and Y. Yamamoto. Functional stochastic reso-nance in human baroreflex induced by 1/f-type noisy galvanic vestibu-lar stimulation. In S. M. Bezrukov, H. Frauenfelder, and F. Moss,editors, Proc. SPIE 5110, Fluctuations and Noise in Biological, Bio-physical, and Biomedical Systems, volume 5110, page 69. InternationalSociety for Optics and Photonics, 5 2003.[358] Y. Song, J. Crowcroft, and J. Zhang. Automatic epileptic seizuredetection in EEGs based on optimized sample entropy and extremelearning machine. Journal of Neuroscience Methods, 210(2):132–146,9 2012.[359] K. M. Spencer, P. G. Nestor, R. Perlmutter, M. A. Niznikiewicz, M. C.Klump, M. Frumin, M. E. Shenton, and R. W. McCarley. Neuralsynchrony indexes disordered perception and cognition in schizophre-nia. Proceedings of the National Academy of Sciences, 101(49):17288–17293, 12 2004.[360] M. B. Spraker, H. Yu, D. M. Corcos, and D. E. Vaillancourt. Roleof Individual Basal Ganglia Nuclei in Force Amplitude Generation.Journal of Neurophysiology, 98(2):821–834, 8 2007.[361] R. Srebro and P. Malladi. Stochastic resonance of the visually evokedpotential. Physical Review E, 59(3):2566–2570, 3 1999.[362] R. J. St George and R. C. Fitzpatrick. The sense of self-motion, ori-entation and balance explored by vestibular stimulation. The Journalof physiology, 589(Pt 4):807–13, 2 2011.167Bibliography[363] C. J. Stagg and M. A. Nitsche. Physiological Basis of TranscranialDirect Current Stimulation. The Neuroscientist, 17(1):37–53, 2 2011.[364] C. J. Stam, G. Nolte, and A. Daffertshofer. Phase lag index: Assess-ment of functional connectivity from multi channel EEG and MEGwith diminished bias from common sources. Human Brain Mapping,28(11):1178–1193, 11 2007.[365] A. Stanca´k and G. Pfurtscheller. Desynchronization and recovery ofbeta rhythms during brisk and slow self-paced finger movements inman. Neuroscience letters, 196(1-2):21–4, 8 1995.[366] T. Stephan, A. Deutschla¨nder, A. Nolte, E. Schneider, M. Wiesmann,T. Brandt, and M. Dieterich. Functional MRI of galvanic vestibularstimulation with alternating currents at different frequencies. Neu-roImage, 26(3):721–732, 7 2005.[367] M. B. Sterman, D. A. Kaiser, and B. Veigel. Spectral analysis of event-related EEG responses during short-term memory performance. BrainTopography, 9(1):21–30, 1996.[368] L. Stiles and P. F. Smith. The vestibularbasal ganglia connection:Balancing motor control. Brain Research, 1597:180–188, 2 2015.[369] L. Stiles, Y. Zheng, and P. F. Smith. The effects of electrical stimu-lation of the peripheral vestibular system on neurochemical release inthe rat striatum. PloS one, 13(10):e0205869, 2018.[370] D. Stoffers, J. L. W. Bosboom, J. B. Deijen, E. C. Wolters, H. W.Berendse, and C. J. Stam. Slowing of oscillatory brain activity is astable characteristic of Parkinson’s disease without dementia. Brain,130(7):1847–1860, 5 2007.[371] F. M. Stoll, C. R. Wilson, M. C. Faraut, J. Vezoli, K. Knoblauch, andE. Procyk. The Effects of Cognitive Control and Time on Frontal BetaOscillations. Cerebral Cortex, 26(4):1715–1732, 4 2016.[372] D. Stru¨ber, S. Rach, T. Neuling, and C. S. Herrmann. On the possiblerole of stimulation duration for after-effects of transcranial alternatingcurrent stimulation. Frontiers in Cellular Neuroscience, 9:311, 8 2015.[373] M. K. Sunwoo, K. H. Cho, J. Y. Hong, J. E. Lee, Y. H. Sohn, and P. H.Lee. Thalamic volume and related visual recognition are associated168Bibliographywith freezing of gait in non-demented patients with Parkinson’s dis-ease. Parkinsonism & Related Disorders, 19(12):1106–1109, 12 2013.[374] M. Tahmasian, L. M. Bettray, T. van Eimeren, A. Drzezga, L. Tim-mermann, C. R. Eickhoff, S. B. Eickhoff, and C. Eggers. A systematicreview on the applications of resting-state fMRI in Parkinson’s disease:Does dopamine replacement therapy play a role? Cortex, 73:80–105,12 2015.[375] K. Tanaka, M. Kawakatsu, and I. Nemoto. Stochastic resonance inauditory steady-state responses in a magnetoencephalogram. ClinicalNeurophysiology, 119(9):2104–2110, 9 2008.[376] L. M. Teles-Grilo Ruivo and J. R. Mellor. Cholinergic modulation ofhippocampal network function. Frontiers in Synaptic Neuroscience,5:2, 2013.[377] C. E. Tenke and J. Kayser. Surface Laplacians (SL) and phase prop-erties of EEG rhythms: Simulated generators in a volume-conductionmodel. International Journal of Psychophysiology, 97(3):285–298, 92015.[378] M. A. Thenganatt and J. Jankovic. Parkinson Disease Subtypes.JAMA Neurology, 71(4):499, 4 2014.[379] W. Thevathasan, P. Mazzone, A. Jha, A. Djamshidian, M. Dileone,V. Di Lazzaro, and P. Brown. Spinal cord stimulation failed to re-lieve akinesia or restore locomotion in Parkinson disease. Neurology,74(16):1325–7, 4 2010.[380] S. Thobois, B. Ballanger, P. Baraduc, D. Le Bars, F. Lavenne,E. Broussolle, and M. Desmurget. Functional anatomy of motor ur-gency. NeuroImage, 37(1):243–252, 8 2007.[381] G. Thut, T. O. Bergmann, F. Fro¨hlich, S. R. Soekadar, J.-S. Brit-tain, A. Valero-Cabre´, A. T. Sack, C. Miniussi, A. Antal, H. R. Sieb-ner, U. Ziemann, and C. S. Herrmann. Guiding transcranial brainstimulation by EEG/MEG to interact with ongoing brain activityand associated functions: A position paper. Clinical Neurophysiol-ogy, 128(5):843–857, 5 2017.[382] G. Thut, P. G. Schyns, and J. Gross. Entrainment of perceptuallyrelevant brain oscillations by non-invasive rhythmic stimulation of thehuman brain. Frontiers in psychology, 2:170, 2011.169Bibliography[383] C. L. Tomlinson, R. Stowe, S. Patel, C. Rick, R. Gray, and C. E.Clarke. Systematic review of levodopa dose equivalency reporting inParkinson’s disease. Movement disorders : official journal of the Move-ment Disorder Society, 25(15):2649–53, 11 2010.[384] S. Tran, M. Shafiee, C. B. Jones, S. Garg, S. Lee, E. P. Pasman, M. G.Carpenter, and M. J. McKeown. Subthreshold stochastic vestibularstimulation induces complex multi-planar effects during standing inParkinson’s disease. Brain Stimulation, 11(5):1180–1182, 9 2018.[385] R. Tremblay, S. Lee, and B. Rudy. GABAergic Interneurons in theNeocortex: From Cellular Properties to Circuits. Neuron, 91(2):260–292, 7 2016.[386] Y. Tufail, A. Matyushov, N. Baldwin, M. L. Tauchmann, J. Georges,A. Yoshihiro, S. I. H. Tillery, and W. J. Tyler. Transcranial PulsedUltrasound Stimulates Intact Brain Circuits. Neuron, 66(5):681–694,6 2010.[387] D. Twelves, K. S. Perkins, and C. Counsell. Systematic review of inci-dence studies of Parkinson’s disease. Movement Disorders, 18(1):19–31, 1 2003.[388] W. J. Tyler. Noninvasive Neuromodulation with Ultrasound? A Con-tinuum Mechanics Hypothesis. The Neuroscientist, 17(1):25–36, 22011.[389] S. Tyll, E. Budinger, and T. Noesselt. Thalamic influences on multi-sensory integration. Communicative & integrative biology, 4(4):378–81,7 2011.[390] P. J. Uhlhaas, C. Haenschel, D. Nikolic´, and W. Singer. The role ofoscillations and synchrony in cortical networks and their putative rele-vance for the pathophysiology of schizophrenia. Schizophrenia bulletin,34(5):927–43, 9 2008.[391] R. J. Uitti, Y. Baba, N. R. Whaley, Z. K. Wszolek, and J. D.Putzke. Parkinson disease: Handedness predicts asymmetry. Neu-rology, 64(11):1925–1930, 6 2005.[392] K. S. Utz, V. Dimova, K. Oppenla¨nder, and G. Kerkhoff. Electri-fied minds: Transcranial direct current stimulation (tDCS) and Gal-vanic Vestibular Stimulation (GVS) as methods of non-invasive brain170Bibliographystimulation in neuropsychologyA review of current data and futureimplications. Neuropsychologia, 48(10):2789–2810, 2010.[393] K. S. Utz, K. Korluss, L. Schmidt, A. Rosenthal, K. Oppenla¨nder,I. Keller, and G. Kerkhoff. Minor adverse effects of galvanic vestibularstimulation in persons with stroke and healthy individuals. BrainInjury, 25(11):1058–1069, 10 2011.[394] V. A. Vakorin, S. M. Doesburg, L. d. Costa, R. Jetly, E. W. Pang, andM. J. Taylor. Detecting Mild Traumatic Brain Injury Using RestingState Magnetoencephalographic Connectivity. PLoS ComputationalBiology, 12(12), 2016.[395] F. Valentino, G. Cosentino, F. Brighina, N. G. Pozzi, G. Sandrini,B. Fierro, G. Savettieri, M. D’Amelio, and C. Pacchetti. Transcranialdirect current stimulation for treatment of freezing of gait: A cross-over study. Movement Disorders, 29(8):1064–1069, 7 2014.[396] S. K. Van Den Eeden, C. M. Tanner, A. L. Bernstein, R. D. Fross,A. Leimpeter, D. A. Bloch, and L. M. Nelson. Incidence of Parkin-son’s disease: variation by age, gender, and race/ethnicity. Americanjournal of epidemiology, 157(11):1015–22, 6 2003.[397] M. P. van den Heuvel and H. E. Hulshoff Pol. Exploring the brainnetwork: A review on resting-state fMRI functional connectivity. Eu-ropean Neuropsychopharmacology, 20(8):519–534, 8 2010.[398] O. van der Groen and N. Wenderoth. Transcranial Random NoiseStimulation of Visual Cortex: Stochastic Resonance Enhances CentralMechanisms of Perception. The Journal of Neuroscience, 36(19):5289–5298, 5 2016.[399] A. van der Hoorn, H. Burger, K. L. Leenders, and B. M. de Jong.Handedness correlates with the dominant Parkinson side: A system-atic review and meta-analysis. Movement Disorders, 27(2):206–210, 22012.[400] E. Van der Ouderaa, J. Schoukens, and J. Renneboog. Peak factorminimization of input and output signals of linear systems. IEEETransactions on Instrumentation and Measurement, 37(2):207–212, 61988.171Bibliography[401] E. Van der Ouderaa, J. Schoukens, and J. Renneboog. Peak fac-tor minimization using a time-frequency domain swapping algorithm.IEEE Transactions on Instrumentation and Measurement, 37(1):145–147, 3 1988.[402] J. Van Doren, B. Langguth, and M. Schecklmann. Electroencephalo-graphic Effects of Transcranial Random Noise Stimulation in the Au-ditory Cortex. Brain Stimulation, 7(6):807–812, 11 2014.[403] B. C. M. van Wijk, P. J. Beek, and A. Daffertshofer. Neural synchronywithin the motor system: what have we learned so far? Frontiers inHuman Neuroscience, 6:252, 2012.[404] J. H. Villafan˜e, K. Valdes, R. Buraschi, M. Martinelli, L. Bissolotti,and S. Negrini. Reliability of the Handgrip Strength Test in ElderlySubjects With Parkinson Disease. Hand (New York, N.Y.), 11(1):54–8, 3 2016.[405] J. L. Vitek, J. Ashe, M. R. DeLong, and G. E. Alexander. Physio-logic properties and somatotopic organization of the primate motorthalamus. Journal of Neurophysiology, 71(4):1498–1513, 4 1994.[406] J. Voges, R. Hilker, K. Bo¨tzel, K. L. Kiening, M. Kloss, A. Kupsch,A. Schnitzler, G.-H. Schneider, U. Steude, G. Deuschl, and M. O.Pinsker. Thirty days complication rate following surgery performedfor deep-brain-stimulation. Movement Disorders, 22(10):1486–1489, 72007.[407] U. Voss, R. Holzmann, A. Hobson, W. Paulus, J. Koppehele-Gossel,A. Klimke, and M. A. Nitsche. Induction of self awareness in dreamsthrough frontal low current stimulation of gamma activity. NatureNeuroscience, 17(6):810–812, 5 2014.[408] A. Vossen, J. Gross, and G. Thut. Alpha Power Increase AfterTranscranial Alternating Current Stimulation at Alpha Frequency (α-tACS) Reflects Plastic Changes Rather Than Entrainment. Brainstimulation, 8(3):499–508, 2015.[409] J. Vosskuhl, D. Stru¨ber, and C. S. Herrmann. Non-invasive BrainStimulation: A Paradigm Shift in Understanding Brain Oscillations.Frontiers in Human Neuroscience, 12:211, 5 2018.172Bibliography[410] M. Vreugdenhil, J. G. R. Jefferys, M. R. Celio, and B. Schwaller.Parvalbumin-Deficiency Facilitates Repetitive IPSCs and GammaOscillations in the Hippocampus. Journal of Neurophysiology,89(3):1414–1422, 3 2003.[411] K. Wakabayashi, K. Tanji, F. Mori, and H. Takahashi. The Lewybody in Parkinson’s disease: molecules implicated in the formationand degradation of alpha-synuclein aggregates. Neuropathology : offi-cial journal of the Japanese Society of Neuropathology, 27(5):494–506,10 2007.[412] J. Wang, C.-T. Zuo, Y.-P. Jiang, Y.-H. Guan, Z.-P. Chen, J.-D. Xiang,L.-Q. Yang, Z.-T. Ding, J.-j. Wu, and H.-L. Su. 18F-FP-CIT PETimaging and SPM analysis of dopamine transporters in Parkinsonsdisease in various Hoehn & Yahr stages. Journal of Neurology,254(2):185–190, 2 2007.[413] L. M. Ward, S. E. MacLean, and A. Kirschner. Stochastic Reso-nance Modulates Neural Synchronization within and between CorticalSources. PLoS ONE, 5(12):e14371, 12 2010.[414] D. L. Wardman, B. L. Day, and R. C. Fitzpatrick. Position and ve-locity responses to galvanic vestibular stimulation in human subjectsduring standing. The Journal of physiology, 547(Pt 1):293–9, 2 2003.[415] G. S. Watson and J. B. Leverenz. Profile of Cognitive Impairment inParkinson’s Disease. Brain Pathology, 20(3):640–645, 5 2010.[416] R. S. Weil, A. E. Schrag, J. D. Warren, S. J. Crutch, A. J. Lees,and H. R. Morris. Visual dysfunction in Parkinsons disease. Brain,139(11):2827–2843, 11 2016.[417] M. Weinberger, W. D. Hutchison, and J. O. Dostrovsky. Pathologi-cal subthalamic nucleus oscillations in PD: Can they be the cause ofbradykinesia and akinesia? Experimental Neurology, 219(1):58–61, 92009.[418] M. Weinberger, N. Mahant, W. D. Hutchison, A. M. Lozano, E. Moro,M. Hodaie, A. E. Lang, and J. O. Dostrovsky. Beta oscillatory activityin the subthalamic nucleus and its relation to dopaminergic responsein Parkinson’s disease. Journal of neurophysiology, 96(6):3248–56, 122006.173Bibliography[419] L. A. Wheaton, H. Shibasaki, and M. Hallett. Temporal activationpattern of parietal and premotor areas related to praxis movements.Clinical Neurophysiology, 116(5):1201–1212, 5 2005.[420] T. Wichmann, M. R. DeLong, J. Guridi, and J. A. Obeso. Milestonesin research on the pathophysiology of Parkinson’s disease. Movementdisorders, 26(6):1032–41, 5 2011.[421] R. Wijesinghe, D. A. Protti, and A. J. Camp. Vestibular Interactionsin the Thalamus. Frontiers in Neural Circuits, 9:79, 12 2015.[422] D. Wilkinson, H. J. Ferguson, and A. Worley. Galvanic vestibularstimulation modulates the electrophysiological response during faceprocessing. Visual neuroscience, 29(4-5):255–62, 9 2012.[423] D. Wilkinson, P. Ko, P. Kilduff, R. McGlinchey, and W. Milberg. Im-provement of a face perception deficit via subsensory galvanic vestibu-lar stimulation. Journal of the International Neuropsychological Soci-ety : JINS, 11(7):925–9, 11 2005.[424] D. Wilkinson, S. Nicholls, C. Pattenden, P. Kilduff, and W. Milberg.Galvanic vestibular stimulation speeds visual memory recall. Experi-mental Brain Research, 189(2):243–248, 8 2008.[425] D. Wilkinson, O. Zubko, M. Sakel, S. Coulton, T. Higgins, and P. Pul-licino. Galvanic vestibular stimulation in hemi-spatial neglect. Fron-tiers in Integrative Neuroscience, 8:4, 1 2014.[426] D. Williams, A. Ku¨hn, A. Kupsch, M. Tijssen, G. van Bruggen,H. Speelman, G. Hotton, C. Loukas, and P. Brown. The relation-ship between oscillatory activity and motor reaction time in theparkinsonian subthalamic nucleus. European Journal of Neuroscience,21(1):249–258, 1 2005.[427] D. Williams, M. Tijssen, G. Van Bruggen, A. Bosch, A. Insola,V. Di Lazzaro, P. Mazzone, A. Oliviero, A. Quartarone, H. Speelman,and P. Brown. Dopamine-dependent changes in the functional con-nectivity between basal ganglia and cerebral cortex in humans. Brain: a journal of neurology, 125(Pt 7):1558–69, 7 2002.[428] J. Wills, J. Jones, T. Haggerty, V. Duka, J. N. Joyce, and A. Sidhu.Elevated tauopathy and alpha-synuclein pathology in postmortemParkinson’s disease brains with and without dementia. Experimen-tal Neurology, 225(1):210–218, 9 2010.174Bibliography[429] B. Wingeier, T. Tcheng, M. M. Koop, B. C. Hill, G. Heit, and H. M.Bronte-Stewart. Intra-operative STN DBS attenuates the prominentbeta rhythm in the STN in Parkinson’s disease. Experimental Neurol-ogy, 197(1):244–251, 1 2006.[430] T. Wu, X. Long, L. Wang, M. Hallett, Y. Zang, K. Li, and P. Chan.Functional connectivity of cortical motor areas in the resting state inParkinson’s disease. Human Brain Mapping, 32(9):1443–1457, 9 2011.[431] T. Wu, L. Wang, Y. Chen, C. Zhao, K. Li, and P. Chan. Changesof functional connectivity of the motor network in the resting state inParkinson’s disease. Neuroscience Letters, 460(1):6–10, 8 2009.[432] Y. Yamamoto, Z. R. Struzik, R. Soma, K. Ohashi, and S. Kwak.Noisy vestibular stimulation improves autonomic and motor respon-siveness in central neurodegenerative disorders. Annals of Neurology,58(2):175–181, 7 2005.[433] C.-G. Yan, X.-D. Wang, X.-N. Zuo, and Y.-F. Zang. DPABI: DataProcessing & Analysis for (Resting-State) Brain Imaging. Neuroinfor-matics, 14(3):339–351, 7 2016.[434] Yi-Ou Li, T. Adali, Wei Wang, and V. Calhoun. Joint Blind SourceSeparation by Multiset Canonical Correlation Analysis. IEEE Trans-actions on Signal Processing, 57(10):3918–3929, 10 2009.[435] H. H. Yin and B. J. Knowlton. The role of the basal ganglia in habitformation. Nature Reviews Neuroscience, 7(6):464–476, 6 2006.[436] T. Zaehle, S. Rach, and C. S. Herrmann. Transcranial AlternatingCurrent Stimulation Enhances Individual Alpha Activity in HumanEEG. PLoS ONE, 5(11):e13766, 11 2010.[437] R. Zink, S. F. Bucher, A. Weiss, T. Brandt, and M. Dieterich. Effectsof galvanic vestibular stimulation on otolithic and semicircular canaleye movements and perceived vertical. Electroencephalography andclinical neurophysiology, 107(3):200–5, 9 1998.[438] R. Zink, S. Steddin, A. Weiss, T. Brandt, and M. Dieterich. Galvanicvestibular stimulation in humans: effects on otolith function in roll.Neuroscience letters, 232(3):171–4, 9 1997.175