You may notice some images loading slow across the Open Collections website. Thank you for your patience as we rebuild the cache to make images load faster.

Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

3D biomechanical oropharyngeal model for training and diagnosis of dysphagia Farazi, Md Moshiur Rahman 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2016_february_farazi_moshiur.pdf [ 18.71MB ]
JSON: 24-1.0220865.json
JSON-LD: 24-1.0220865-ld.json
RDF/XML (Pretty): 24-1.0220865-rdf.xml
RDF/JSON: 24-1.0220865-rdf.json
Turtle: 24-1.0220865-turtle.txt
N-Triples: 24-1.0220865-rdf-ntriples.txt
Original Record: 24-1.0220865-source.json
Full Text

Full Text

3D Biomechanical OropharyngealModel for Training and Diagnosis ofDysphagiabyMd Moshiur Rahman FaraziB.Sc., Islamic University of Technology, 2011M.Sc., Islamic University of Technology, 2013A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF APPLIED SCIENCEinThe Faculty of Graduate and Postdoctoral Studies(Biomedical Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)December 2015c© Md Moshiur Rahman Farazi 2015AbstractSwallowing is a complex oropharyngeal process governed by intricate neu-romuscular functions. Dysfunction in swallowing, clinically termed as dys-phagia, can significantly reduce the quality of life. Modified barium swallow(MBS) studies are performed to produce vidoefluoroscopy (VF) for visual-izing swallowing dynamics to diagnose dysphagia. To train the clinicianslearning standardized dysphagia diagnosis, 2D animated videos coupled withVF are used. However, it is hypothesised that the physiologic components ofthe oral domain may benefit from extension of the training materials, suchas inclusion of 3D models.We develop a 3D biomechanical swallowing model of the oropharyngealcomplex to extend the clinical dysphagia diagnosis training materials. Ourapproach incorporates realistic geometries and accurate timing of swallowingevents derived from training animations that have been clinically validated.We develop rigid body models for the bony structures and finite element mod-els (FEM) for the deformable soft structures, and drive our coupled biome-chanical model kinematically with accurate timing of swallowing events. Weimplement an airway-skin mesh using a geometric skinning technique thatunifies geometric blending for rigid body model with embedded surface forFEMs to incorporate the deformation of upper airway during a swallowingmotion. We use smoothed particle hydrodynamics (SPH) technique to sim-ulate a fluid bolus in the airway-skin mesh where the model dynamics drivethe bolus to emulate bolus transport during a swallowing motion.We validate this model in two phases. Firstly, we compare the simulatedbolus movement with input data and match the swallowing kinematics iden-tified in the standardized animations. Secondly, we extend existing trainingmaterial for standardized dysphagia diagnosis with our 3D model. To test theusefulness of the extended training set using 3D visualizations, we conducta pilot user study involving Speech Language Pathologists. The pilot dataindicate that clinicians believe the additional 3D views are useful for iden-tifying the salient features for differentiating between different swallowingiiAbstractimpairments, such as direction, strength and timing of the tongue motion,and could be a useful addition to the current standardized MBSImPTM c©training system.iiiPrefaceParts of this dissertation have been published elsewhere.Full Length PaperVersions of Chapter 3 and 4 have been published in• Farazi, M.R., Martin-Harris, B., Harandi, N., Fels, S. and Abughar-bieh, R. “A 3D Dynamic Biomechanical Swallowing Model for Trainingand Diagnosis of Dysphagia”, International Symposium on BiomedicalImaging (ISBI), Brooklyn-USA, April Pages: 1385–1388, 2015.MR Farazi was the main contributor of this paper under the supervision ofDr. Rafeef Abugharbieh and Dr. Sidney Fels. MR Farazi developed the 3Doropharyngeal model, performed simulations and analysis, generated figuresand results, and prepared the manuscript for publication. Dr. Martin-Harrisprovided the VF and animation data, and helped to formulate the researchquestion with clinical significance. N Harandi helped prepare the manuscript.Dr. Sidney Fels and Dr. Rafeef Abugharbieh carefully edited the manuscriptand provided additional feedback.Extended AbstractsSome parts of Chapter 3 have been published as extended abstracts whichare listed below:• Farazi, M.R., Martin-Harris, B., Abugharbieh, R. and Fels, S. “Devel-opment of a 3D Biomechanical Swallowing Model for Dysphagia Train-ing”, International Symposium on Computer Methods in Biomechanicsand Biomedical Engineering (CMBBE), Montreal-Canada, September2015ivEthics Applications• Farazi, M.R., Martin-Harris, B., Abugharbieh, R. and Fels, S. “Swal-lowing Simulation Using an MBSImP-Based 3D Biomechanical Model”,International Symposium on Computer Methods in Biomechanics andBiomedical Engineering (CMBBE), Amsterdam-Netherlands, October2014.For both publications, MR Farazi was the primary author and main contrib-utor to the design, implementation, and testing of the methods developed,under the supervision of Dr. Rafeef Abugharbieh and Dr. Sidney Fels. Dr.Martin-Harris provided the VF and animation data and helped with theanalysis. M Farazi prepared the manuscript for the publication which wascarefully reviewed and edited by Dr. Rafeef Abugharbieh and Dr. SidneyFels.Finally, a combination of Chapters 3, 4 and 5 will be submitted as a fulljournal article which is now under preparation.Ethics ApplicationsThis research met the minimal risk human ethics application criteria andthus an expedited review was conducted for both of the Behavioural Re-search Ethics Board (BREB) and Clinical Research Ethics Board (CREB)applications, required to conduct this research.• The videofluoroscopic data used in this Thesis was provided by Dr.Bonnie Martin-Harris of Medical University of South Carolina (MUSC).This data sharing was approved by UBC Clinical Research EthicsBoard (CREB) application number H15-02749.• The user study conducted in this Thesis was approved by UBC Be-havioural Research Ethics Board (BREB) application number H15-02665.Conflict of InterestThe researchers and members of the thesis committee report no conflict ofinterest.vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . xiiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xivDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Problem Statement and Scope . . . . . . . . . . . . . . . . . 41.3 Physiology of a Normal Swallowing . . . . . . . . . . . . . . . 41.3.1 Oral preparatory phase . . . . . . . . . . . . . . . . . 61.3.2 Oral phase . . . . . . . . . . . . . . . . . . . . . . . . 61.3.3 Pharyngeal phase . . . . . . . . . . . . . . . . . . . . 91.3.4 Esophageal phase . . . . . . . . . . . . . . . . . . . . 101.3.5 Airway protection mechanism . . . . . . . . . . . . . . 121.4 Diagnostic Evaluation of Dysphagia . . . . . . . . . . . . . . 121.4.1 Clinical bedside swallow assessment . . . . . . . . . . 131.4.2 Modified Barium Swallow Study (MBSS) . . . . . . . 151.4.3 Endoscopic Evaluation of Swallowing . . . . . . . . . . 161.4.4 Manometry . . . . . . . . . . . . . . . . . . . . . . . . 181.4.5 Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . 20viTable of Contents1.5 Clinical Decision Making . . . . . . . . . . . . . . . . . . . . 211.6 Standardized Evaluation of Dysphagia Diagnosis . . . . . . . 221.6.1 Modified Barium Swallow Impairment Profile . . . . . 231.6.2 Reliability training and testing . . . . . . . . . . . . . 241.7 Biomechanical Model in Standardized Dysphagia Training . . 251.8 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 251.9 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.1 Resources Used for Clinical Education & Training . . . . . . 282.2 Mannequin Simulator Models . . . . . . . . . . . . . . . . . . 292.3 Animation Models . . . . . . . . . . . . . . . . . . . . . . . . 312.4 Biomechanical Simulation Models . . . . . . . . . . . . . . . . 342.4.1 Oropharyngeal geometries . . . . . . . . . . . . . . . . 342.4.2 Simulation of bolus . . . . . . . . . . . . . . . . . . . 352.4.3 Grid Based Methods . . . . . . . . . . . . . . . . . . . 352.4.4 Meshfree methods . . . . . . . . . . . . . . . . . . . . 382.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 Modeling of the Oropharyngeal Complex . . . . . . . . . . . 423.1 Building Geometries . . . . . . . . . . . . . . . . . . . . . . . 423.1.1 Generating full 3D model . . . . . . . . . . . . . . . . 433.1.2 Realistic morphometry . . . . . . . . . . . . . . . . . . 463.2 Building the biomechanical model . . . . . . . . . . . . . . . 483.2.1 Extracting kinematics from animation . . . . . . . . . 483.2.2 Coupled biomechanical model . . . . . . . . . . . . . . 493.2.3 Adding kinematics . . . . . . . . . . . . . . . . . . . . 523.3 Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . 533.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Bolus Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 574.1 Deformable Airway Model . . . . . . . . . . . . . . . . . . . . 574.1.1 Airway geometry . . . . . . . . . . . . . . . . . . . . . 584.1.2 Deforming airway-skin with model dynamics . . . . . 594.2 Bolus Simulation Using SPH . . . . . . . . . . . . . . . . . . 614.2.1 Artificial viscosity term . . . . . . . . . . . . . . . . . 634.2.2 Artificial compressibility . . . . . . . . . . . . . . . . . 644.2.3 Time step selection . . . . . . . . . . . . . . . . . . . 66viiTable of Contents4.2.4 Boundary treatment . . . . . . . . . . . . . . . . . . . 674.3 Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . 704.3.1 Simulation of a stable bolus . . . . . . . . . . . . . . . 704.3.2 Simulating bolus with standardized consistencies . . . 724.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735 Extending Standardized Dysphagia Training Materials . . 745.1 Clinical Collaborators . . . . . . . . . . . . . . . . . . . . . . 745.2 Study Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 745.3 Phase I- Interviewing the Stakeholders . . . . . . . . . . . . . 755.4 Phase II - User Study . . . . . . . . . . . . . . . . . . . . . . 785.4.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . 795.4.2 Participant selection criteria . . . . . . . . . . . . . . 815.4.3 User interface . . . . . . . . . . . . . . . . . . . . . . . 815.5 Results and analysis . . . . . . . . . . . . . . . . . . . . . . . 865.5.1 Questionnaire analysis . . . . . . . . . . . . . . . . . . 865.5.2 Testimonial data . . . . . . . . . . . . . . . . . . . . . 895.5.3 Recommendation . . . . . . . . . . . . . . . . . . . . . 905.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 925.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 926 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 946.1 Impact and Potential Application . . . . . . . . . . . . . . . . 946.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . 96AppendicesA MBSImPTM c© scoring guideline . . . . . . . . . . . . . . . . . 99A.1 MBSImPTM c© Components, Scores, and Score Definitions . . 99B Formulation for SPH fluid simulation . . . . . . . . . . . . . 101B.1 Integral representation of a function . . . . . . . . . . . . . . 101B.2 Integral representation of the derivative of a function . . . . . 104B.3 Particle approximation . . . . . . . . . . . . . . . . . . . . . . 107B.4 Choice of smoothing function . . . . . . . . . . . . . . . . . . 109B.5 SPH equation of motion . . . . . . . . . . . . . . . . . . . . . 110B.6 The continuity equation . . . . . . . . . . . . . . . . . . . . . 112viiiTable of ContentsB.7 The momentum equation . . . . . . . . . . . . . . . . . . . . 113B.8 The energy equation . . . . . . . . . . . . . . . . . . . . . . . 113B.9 Navier-Stokes equations . . . . . . . . . . . . . . . . . . . . . 114B.10 Particle approximation of density . . . . . . . . . . . . . . . . 115B.11 Particle approximation of momentum . . . . . . . . . . . . . 116B.12 Particle approximation of energy . . . . . . . . . . . . . . . . 117Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118ixList of Tables1.1 Clinical specialties involved in dysphagia diagnosis . . . . . . . 21.2 List of MBSImPTM c© physiologic components . . . . . . . . . 242.1 Comparison of Lagrangian and Eulerian grid methods . . . . . 373.1 Neonatal and adult tongue dimensions . . . . . . . . . . . . . 47B.1 Particle approximation representation of the value of a func-tion and its derivatives at particle i . . . . . . . . . . . . . . . 109xList of Figures1.1 Human upper airway anatomy, VF image and illustration . . . 51.2 Different phases of a swallowing motion, VF image and illus-tration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71.3 Muscles of the tongue . . . . . . . . . . . . . . . . . . . . . . . 81.4 Time–line of oral and pharyngeal swallowing phase . . . . . . 91.5 Peristalsis during esophageal phase of swallowing . . . . . . . 111.6 A modern videofluoroscopy machine. . . . . . . . . . . . . . . 141.7 Normal swallowing motion viewed during a MBS study . . . . 151.8 A Fiberoptic Endoscopic Evaluation of Swallowing (FEES)system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171.9 Endoscopic view of epiglottis, vocal cords, trachea, and cartilage 181.10 Schematic for esophageal manometry . . . . . . . . . . . . . . 191.11 Ultrasound evaluation of tongue motion . . . . . . . . . . . . . 201.12 Operational definition of “Component 6 - initiation of pharyn-geal swallow” of MBSImPTM c© . . . . . . . . . . . . . . . . . 232.1 Mannequin simulator models . . . . . . . . . . . . . . . . . . . 292.2 Animated illustration used in standardized dysphagia training 312.3 Animation training models . . . . . . . . . . . . . . . . . . . . 322.4 FEES training with simulator models . . . . . . . . . . . . . . 333.1 Animated swallowing model of the oropharyngeal complex . . 443.2 Animated illustration of swallowing motion . . . . . . . . . . . 453.3 Building full 3D geometries of oropharyngeal complex . . . . . 463.4 Dimension of the tongue in the model . . . . . . . . . . . . . . 483.5 Adding kinematic to the 3D model . . . . . . . . . . . . . . . 493.6 Components of the 3D oropharyngeal model . . . . . . . . . . 503.7 Generating FEM from oropharyngeal surface geometries . . . 513.8 Kinematically driven model and animated illustration of swal-lowing motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 53xiList of Figures3.9 Kinematically driven model and videofluoroscopic image ofswallowing motion . . . . . . . . . . . . . . . . . . . . . . . . 554.1 Coupling of airway-skin mesh . . . . . . . . . . . . . . . . . . 594.2 Deformation of the airway-skin mesh . . . . . . . . . . . . . . 604.3 Real and virtual SPH particles . . . . . . . . . . . . . . . . . . 684.4 Swallowing simulation results compared with VF and stan-dardized animation video . . . . . . . . . . . . . . . . . . . . . 715.1 Existing MBSImPTM c© online training platform . . . . . . . . 775.2 Tongue motion for scoring a 3 for Component 4 . . . . . . . . 805.3 Online training interface for the user study . . . . . . . . . . . 825.4 Component score 4-0 with additional 3D views . . . . . . . . . 835.5 Component score 4-3 with additional 3D views . . . . . . . . . 845.6 Variation of age and experience in the study population . . . . 865.7 Median and Mode of the questionnaire response score . . . . . 875.8 Average–maximum–minimum chart for the questionnaire re-sponse score . . . . . . . . . . . . . . . . . . . . . . . . . . . . 885.9 Evaluation of Component 15 – tongue base retraction . . . . . 91B.1 Support domain and problem domain for smoothing functionW with surface integral equal to zero . . . . . . . . . . . . . . 106B.2 Support domain and problem domain for smoothing functionW with surface integral not equal to zero . . . . . . . . . . . . 106xiiList of AbbreviationsList of AbbreviationsDOF Degree of FreedomFDM Finite-Difference MethodFEES Fiberoptic Endoscopic Evaluation of SwallowingFEM Finite Element MethodSLP Speech-Language PathologistSPH Smoothed Particle HydrodynamicsVF VideofluoroscopyVFSS Videofluoroscopic Swallowing StudyxiiiAcknowledgementsI am thankful to Almighty Allah, most Gracious, who in His infinite mercyhas guided me to complete this dissertation. Foremost, I would like to expressmy sincere gratitude to my supervisor Prof. Rafeef Abugharbieh, whoseexpertise, understanding and patience added considerably to my graduateexperience. I would like to express my sincere gratitude to my co-supervisorProf. Sidney Fels for his continuous support throughout my graduate study.His guidance, mentorship, patience and motivation helped me significantlyduting this research and preparing this manuscript.Beside my supervisors, I would like to thank our clinical collaborator Dr.Bonnie Martin-Harris of Medical University of South Carolina (MUSC), SC,USA for the productive collaboration over the past years. Her patience,enthusiasm, appreciation and immense knowledge motivated me to workharder. Without her support it would not have been possible to conductthis research.I thank my fellow researchers at Biomedical Signal and Image ComputingLaboratory (BiSICL) and Human Communication Technologies (HCT) Labat UBC for all the stimulating discussions and for all the fun we have had inthe last couple of years.Finally, a heart felt thank-you to my parents and family members forproviding me with a loving environment that let me learn and grow. I amgrateful for their sacrifices, encouragement and support throughout my life.xivDedicationDedicated to my parents and my familyxvChapter 1Introduction1.1 MotivationSwallowing is a physiologic process by which food travels from the mouth,through the pharynx and into the esophagus. Swallowing has two crucialbiological features: food passage from the oral cavity to stomach and airwayprotection. Although it seems quite simple, swallowing is a very complexneuromuscular function involving voluntary and reflexive activities of morethan 30 nerves and muscles[1]. Difficulty or inability to swallow is clinicallytermed as dysphagia which is derived from the Greek words dys meaningbad, disordered, impaired and the root phag meaning “eat”. So dysphagiameans difficulty or discomfort in swallowing.Dysphagia can occur in any age but is most prevalent in the elderlypopulation. Epidemiological data estimates that prevalence of dysphagiaamong people over 50 years can be as high as 22% [2; 3]. Dysphagia can leadto dehydration, malnutrition as it impedes the normal nutrition required inthe body. As a result, patients suffering from dysphagia are at high risk ofdeveloping other medical conditions. In more severe cases of dysphagia, theairway protection mechanism fails and food enter the lungs through trachearather than going to stomach through esophagus, which is called aspirationpneumonia. Aspiration pneumonia is a major cause leading to hospitalizationfor elderly people which causes costly care, and overall mortality rate rangesform 20% to 50% [4; 5; 6; 7], if not treated. Furthermore, dysphagia addsto the patients suffering as it has significant social and physiological impact[8]. For example, eating and drinking are most common social activity inalmost every society. Often meals are the main activity of celebrations andgatherings. So dysphagia can sometimes cause social isolation, jeopardizeones social experience and degrade quality of life significantly.Dysphagia can be a result of abnormalities related to neural control, mus-cle contraction, inflammation and commonly associated symptom of stroke,multiple sclerosis (MS), cancer and many other diseases that impede normal11.1. MotivationDentistry GastroenterologyGeneral Surgery Head and Neck surgeryNeurology NeurosurgeryNursing NutritionOccupational Therapy OncologyOral Surgery OtolaryngologyPediatrics Physical TherapyPulmonary Medicine RadiologyRehabilitation RheumatologySpeech-Language Pathology Thoracic SurgeryTable 1.1: Clinical specialties involved in the evaluation and management ofswallowing disordersfunctionalities of oropharyngeal nerves and muscles. Patients with dysphagiarequire a multidisciplinary approach to swallowing management. This mayinclude swallow therapy, dietary modification and oral care. A large anddiverse number of clinical specialties (see Table 1.1) are concerned with theevaluation and managements of dysphagia. Management and treatment ofdysphagia can improve the patients nutrition. Improving patients care andoffering them appropriate treatment could prevent the physiological damageand social isolation that could otherwise happen to the patient.Simulating a swallowing motion itself is of interest of many disciplines(see Table 1.1) and the research is motivated by the application areas of themodels. Some application that require simulation of swallowing motion arelisted below:Food texture modificationPatient suffering from dysphagia are in danger of aspiration pneumonia andthe chances of death due to this is high for elderly patients [2; 3]. To improvethe quality of life for dysphagiac patients, there is a need for development oforopharyngeal swallowing model for developing “safety food” by modifyingthe food texture [9]. Without a swallowing model, the determination ofthe safety food for dysphagic patients involved the testing of different foodconsistencies by trail and error. This method involves the risk of chockingor aspiration which is not desirable. As a result, research is going on fordeveloping a swallowing model for understanding the action of food bolus21.1. Motivationand corresponding oropharyngeal structures during the swallowing process.There are also some parallel research going on aiming to solve the sameproblem by developing robotic swallowing simulators using motors, actuatorsand muscular tube[10].Surgical planningComputer modeling of anatomical structures has proven to be useful in headand neck cancer management and otolaryngology. A significant number ofpatients undergoing head and neck surgery develop swallowing and speechrelated problems. Such computer models are built to predict anatomical andfunctional outcome during surgical planning and demonstrating the prognosisto a patient before the surgery. These models help reduce the treatment timesignificantly and help evaluate the risk of aspiration and other swallow relatedcomplexities. For addressing these issues, research is going on for developingpatient specific computer models of oropharyngeal complex for predictingfunctional mastication and swallowing outcome.Obstructive sleep apnea (OSA)Obstructive sleep apnea (OSA) is a symptom characterized by partial orcompleted obstruction of the airway during sleep. Such obstruction causesnarrowing of the pharynx, which decrease the tone of the pharyngeal dilatormuscle [11] and cause snoring. Some studies [12; 13; 14] suggest that theneurogenic lesions in the pharynx and soft palate are triggered by vibrationsproduced by snoring. The pharynx and soft palate are two major oropha-ryngeal structures that contribute to the propagation of food bolus and suchlesions impair their sensitivity and functionality. It is hypothesized that pa-tients with OSA may have subclinical swallowing abnormalities due to lowfrequency vibrations, and negative intrathoracic pressure during apnoea [14].To explore this cause and effect relationship between OSA and dysphagia,computer models of the oropharangeal complex are developed to simulateOSA. The application of such model include condition by developing oralappliance for OSA patients and so on.Therefore, there exists a need for a biomechanical oropharyngeal swal-lowing model in dysphagia research. The application of such biomechani-cal model include but are not limited to, standardized dysphagia training,post–operative rehabilitation, treatment planning. A survey conducted by31.2. Problem Statement and ScopeAmerican Speech-language Hearing Association (ASHA) reported that 92%and 100% speech-language pathologists (SLPs) working respectively in hos-pital settings and residential health-care setting such as nursing home or carefacilities, regularly serve individuals with swallowing disorders [15]. The de-gree and type of training in diagnosis and evaluation of dysphagia are highlyvaried and sparse [16]. Contribution towards the clinician training to eval-uate and diagnose dysphagia would have tremendous impact on dysphagiaresearch. So in this Thesis, we developed a biomechanical swallowing modelto help clinician training for dysphagia diagnosis.1.2 Problem Statement and ScopeClinician training is instrumental to establish a standardized diagnostic pro-tocol. Such training ensures that clinicians across institutions can score re-liably and communicate the test results across the continuum of care. Inthis Thesis, a biomechanical modeling approach is investigated in order toassist clinicians in standardized training of dysphagia using a 3D biomechan-ical swallowing model. The model geometries are built from geometries andclinically validates timings of swallowing event. The model is built usingrealistic geometries and kinematic information derived from clinical train-ing data used in MBSImPTM c© [17]. With such physics driven models, theclinicians would be able to alter the timing of swallowing movements andvisualize the effect of these simulated interventions on bolus clearance andairway protection. By interactively changing views and altering timing ofswallowing movements, we anticipate that clinicians will gain deeper under-standing of the complex dynamics involved in swallowing, especially as itrelates to different patients, to facilitate diagnosis and treatment planning.1.3 Physiology of a Normal SwallowingAs stated earlier, swallowing is a very complex synchronized action under-taken by oral, pharyngeal and laryngeal muscles and structures (see Figure1.1). During a normal swallow, the bolus is propelled from the mouth throughthe oral cavity, pharynx and esophagus by the positive pressure applied tothe bolus tail. At the beginning of swallowing, the tongue holds the bolus atthe oral cavity by having the tongue tip and base elevated against the front41.3. Physiology of a Normal SwallowingFigure 1.1: Mid-sagittal VF and illustration of oropharyngeal complex. VFimage reproduced from MBSImPTM c© training module, with permission.teeth and soft palate respectively. Following the holding motion, the tonguetip moves along the roof of the mouth and the base separates from the softplate allowing the bolus to flow from the oral cavity into the pharynx. After-words the bolus passes the upper esophageal sphincter (UES) and throughthe esophagus to the stomach for digestion.Swallowing is historically divided in to four phases [18] namely (see Figure1.2):1. Oral preparatory phase,2. Oral phase,3. Pharyngeal phase and4. Esophageal phase.The first two phases are modulated primarily by voluntary control how-ever last two are governed by involuntary control. Martin-Harris et al. [19]reported that oropharyngeal swallow events do not segment into discrete oraland pharyngeal phases and there is a overlap between the initiation of oraland pharyngeal phase of swallow. Some studies [20; 21; 22] investigated theswallowing physiology with manometric and videofluoroscopic analysis anddemonstrated that physiologic events during swallowing are inter-dependent,51.3. Physiology of a Normal Swallowingwhich means, one event impacts another. These findings contradict the tra-ditional description of swallowing mechanism which is based on visual ob-servations and temporal measurement of structural movements and bolusflow throughout the upper aerodigestive tract from lateral videofluoroscopicrecordings. However, having the complex swallowing mechanism divided intodiscrete phases and discussing about physiological characteristics individualto that particular phase, makes it easier to differentiate between normal andimpaired swallow. For this reason following subsections discuss about differ-ent phase of swallowion mottion even if these phase are not independent andmay be overlapping.1.3.1 Oral preparatory phaseDuring the oral preparatory phase the food is tasted, broken down by masti-cation and prepared for swallowing. At first the lips are closed off and musclesare activated to contain the liquid and/or the food inside the oral cavity. Thelip closure ensures the seal at the anterior part. On the other hand, the softpalate touches the tongue base to seal off the oral cavity posteriorly . Theposterior seal is an important airway protection mechanism as it preventspremature slippage of food or liquid into the oropharynx [23]. Masticationtakes place once the bolus containment is achieved and the jaw performs alateral circular or rotary motion to crush the food into small pieces. Duringmastication of a solid bolus, tongue is moved laterally and vertically to posi-tion the food between the teeth and the bolus is mixed with saliva. Withoutnormal range of tongue movement and muscle control, tongue mobility can-not be achieved which is the most important activity during oral preparatoryphase.1.3.2 Oral phaseAfter the bolus is prepared at the oral preparatory stage, the oral phase(sometimes in literature referred as oral transport stage) begins. The newlyformed cohesive bolus is propelled through the oral cavity to be transportedinto the oropharynx. The tongue propels the bolus posteriorly with a upwardand backward rolling motion while the soft palate is elevated to seal off thenasopharynx. The oral- and nasopharangeal seal is important as they createclosed pressure system within the oral cavity to facilitate the bolus trans-port. This wavelike motion is both centripetal and centrifugal in nature. It61.3. Physiology of a Normal SwallowingFigure 1.2: Lateral view of a healthy swallow during an modified bariumswallow (MBS) exam showing different phases of swallowing motion. VFimage reproduced from MBS video used in MBSImPTM c© training, withpermission.71.3. Physiology of a Normal SwallowingFigure 1.3: (a) Lateral and (b) anterior view of the extrinsic tongue musclesthat insert into the tongue from outside origins and allow the tongue to movein different directions. Images (a) and (b) are adapted respectively from [26]and [27] c licence via Wikimedia Commons.occurs as a result of activity within the intrinsic and extrinsic tongue mus-cles (see Figure 1.3) of the tongue i.e. genioglossus, hyoglossus, styloglossus,palatoglossus, superior longitudinal [24]. The extrinsic tongue muscle nameshave the root word “glossus” = tongue at the end and at the beginning hasthe name of the origin. For example, genioglossus from “genio” = chin, sty-loglossus from styloaid bone, palatoglossus from soft palate, hyoglossus fromhyoid bone. The intrinsic muscles originates within the tongue and allowthe tongue to change its shape. The bolus enters into the oropharynx bydepression of the posterior tongue while the soft palate is elevated. This oralphase of swallow in terminated when the swallowing reflex is triggered. Thenormal duration of this phase is around 1s and it is not varied significantlywith bolus consistency, age, sex of people [25].81.3. Physiology of a Normal SwallowingFigure 1.4: Time line showing timing of events for the oral and pharyn-geal phase of swallow for 5mL barium bolus swallow in normal subject. TB,tongue base movement; SM-O, onset of submental electric activity; TT, onsetof propulsive tongue tip; SH-O, onset of superior hyoid movement; SL-O, on-set of superior laryngeal movement; AH-O, onset of anterior hyoid movement;AL-O, onset of anterior laryngeal movement; SH-C, completion of superiorhyoid movement. Image adapted from [32].1.3.3 Pharyngeal phaseDuring the pharyngeal phase the bolus passes through the pharynx. Thisstage also lasts about 1s and despite its short duration it is the most com-plex of all swallowing phases. It requires quick and precise coordination ofalmost all swallowing musculature. The swallowing reflex is triggered by theglossopharyngeal nerve and it marks the beginning of the pharyngeal phase[28]. Bolus properties e.g. texture, taste, and volume can alter the timing ofthe involuntary “trigger” for the pharyngeal phase. For example, more vis-cous liquids may delay the pharyngeal trigger [29], on the other hand, sourfoods might initiate an earlier trigger [30]. In addition, bolus size can alsoinfluence the timing of the trigger. For example, while drinking large amountof water in a continuous sequence, boluses may pass the faucial pillars andreach the valleculae before the trigger is initiated [31].When the swallowing reflex is triggered some sequential neuromotor ac-91.3. Physiology of a Normal Swallowingtivities occur. In literature, sometimes swallowing is quantified in a time–line(see Figure 1.4) to map the sequential and overlapping swallow related eventswith respect to time [32]. Firstly, when the bolus passes the opening onto thenasal cavity, velopharyngeal closure is achieved to ensure that the nasal cav-ity is closed and no bolus is entering the nose. This happens in a fraction of asecond and respiration ceases during this period to ensure airway protection.Next, the pharyngeal muscles contract and the pharynx is elevated superiorly.The tongue base is retracted towards the posterior pharyngeal wall simul-taneously and the pharyngeal constrictors are activated in a rostral-caudaldirection [25]. As a result, pharyngeal constrictor contracts in a wavelike mo-tion called pharyngeal peristalsis or the pharyngeal stripping wave because itstrips along the tail of bolus and squeezing it through the pharynx and intothe upper esophagus. The pharyngeal stripping wave generates an averagepressure of 22 mmHg and an average force of 1.2 mmHg*s [33]. This wavedescends inferiorly from the level of the nasopharynx to the level of the upperesophageal sphincter (UES) at a rate of between 9 and 25 cm/s [33].Due to the elevation of the pharynx, the suprahyoid muscles contract todirect the hyoid bone superiorly and anteriorly [34]. Consequently, the thyro-hyoid muscle contracts to move the larynx superiorly towards the hyoid bone.Anterior and superior movement of the larynx is important for a number ofreasons. Firstly, these movements direct the larynx under the tongue baseand inverting the epiglottis. It pushes the bolus away from the laryngeal inletwhich is an important part of airway protection mechanism. Secondly, si-multaneous elevation of larynx and hypopharynx creates a negative pressurebelow the level of the bolus. This pressure helps to guide the bolus infe-riorly towards the esophagus. Finally, the elevation of larynx and pharynxcreates a biomechanical force that helps to pull the cricoid cartilage up andaway from the posterior pharyngeal wall. As a result, this force pulls openthe cricopharyngeal muscle and the UES. The opening of the UES createsan additional suction force at the bolus for guiding it towards the esopha-gus. The combined effect and synergy of the suction force, tongue movementand pharyngeal stripping wave determines the efficiency of pharyngeal bolustransit.1.3.4 Esophageal phaseThe esophageal phase of swallowing begins when the bolus is passed throughthe UES. The biomechanical forces contributing to the opening of the UES101.3. Physiology of a Normal SwallowingFigure 1.5: Esophagus contracts in an orderly sequence from top to bot-tom (peristalsis) in order to transport the swallowed bolus into the stomach.Adapted from [35] with c licence via Wikimedia Commons.and relaxation of the cricopharyngeal muscle facilitates the UES opening.The relaxation lasts for about 0.5 to 1.2 seconds which is just enough timefor the bolus to pass through the UES and into the esophagus. The cricopha-ryngeal muscle returns to its contracted state sealing off the esophagus oncethe bolus has successfully entered into the esophagus. This seal prevents anybolus that is directed backwards form entering the hypopharynx. Esophagealperistalsis (see Figure 1.5) is activated at this point and the bolus is pushedtowards the lower esophageal sphincter (LES) and stomach. This esophagealperistaltic wave squeeze the bolus through the esophagus and travels at arate approximately 3 to 4 cm/s [36]. Once the LES is triggered to relax,the peristaltic waves push and squeeze the bolus into the stomach which isfollowed by several secondary waves that can last up to an hour after theswallow. These secondary waves help to clear any remaining bolus residuesin the esophagus [37]. Transit time during this phase is usually 8 to 13s,111.4. Diagnostic Evaluation of Dysphagiahowever, this time can vary with age, bolus size and texture [38].1.3.5 Airway protection mechanismAirway protection is an important characteristic of a healthy and normalswallowing motion. People aspirate when airway protection mechanism failsand bolus enters the airway rather than going into the esophagus. Such as-piration, if untreated, can result in life threatening aspiration pneumonia.Proper coordination and timing of swallow events is crucial for ensuring air-way protection as respiration and swallowing share similar anatomical path-ways. Usually, when a person is starting to exhale, pharyngeal phase ofswallow takes place [39; 40]. During swallowing, respiration ceases till thebolus has cleared the hypopharynx and entered the esophagus [39; 40].Vocal folds and epiglottic deflection are important physiological activitythat contributes to airway protection. Closure of the vocal fold is achievedjust before the initiation of pharyngeal phase and it halts the respiration byclosing off the airway. This seal prevents any misdirected foods bolus residuefrom entering the esophagus [41]. Further, during epiglottic deflection, theinferior surface of the epiglottic makes contact with the arytenoid cartilageswhich then directs the bolus past the airway and into the esophagus. As aresult, vocal folds and epiglottic deflection acts as shields for the airway, pre-venting it against penetration and aspiration of liquid. Gag reflex has beenhistorically regarded as a airway protection mechanism. However, absence ofa gag reflex does not necessarily indicate that a patient is unable to swallowsafely. Indeed, many individual with an absent gag reflex have normal swal-lowing, and some patients with dysphagia have a normal gag reflex. Somestudies have also proven that reduced or absent gag reflex is not correlatedwith an increased risk for aspiration [42].1.4 Diagnostic Evaluation of DysphagiaDifferent swallowing assessment tools have been developed to diagnose andtreat dysphagia. A swallowing assessment is a diagnostic procedure aimedto identify anatomy and physiology of all stages of the patients swallow ac-curately and detect any swallow related impairments. To identify swallowingabnormalities, it is important to perform a swallowing assessment, and to ini-tiate early referral for diagnosis and treatment to minimize health risks. The121.4. Diagnostic Evaluation of Dysphagiaclinical tools to diagnose dysphagia can be broadly divided into 2 categories:(1) bedside and (2) instrumental assessment. If any sign of dysphagia is foundduring the bedside tests, which is a physiological assessment, the cliniciansdo a comprehensive instrumental assessment to observe the exact appear-ance and coordination of movement of the oropharyngeal structures and toevaluate aspiration. Instrumental evaluation include imaging techniques likevideofluroscopy, fiberoptic endoscopy, ultrasound and nonimaging techniqueslike manometry to evaluate a swallowing motion. These tests provide addi-tional information about the patients’ swallowing mechanism to help theclinician diagnose the abnormality and devise treatment plan. Some widelyused swallowing assessment techniques are listed in the following sections.1.4.1 Clinical bedside swallow assessmentBedside or initial swallowing assessment is usually the first swallowing assess-ment done when an individual is presented with swallowing discomfort. Onestudy show that, 42% – 60% of acute stroke patients have been reported tohave dysphagia on the basis of standardized clinical bedside within a medianof 3 days from stroke [43]. This bedside test refers to a minimally invasiveevaluation procedure of swallow motion that provides quick determinationof the likelihood that dysphagia exists. Such screening is often performed toknow if someones’ swallowing ability has been compromised after a stroke orsome other neurological diseases.The first step of this bedside swallow assessment (often called a swallowscreen) is to make sure that the patient can follow simple one step commandsto be able to participate in the assessment. The patient is asked to open themouth and the clinician, most often a Speech-Language Pathologist (SLP)with training in the area of evaluation and treatment of dysphagia, looks intothe oral cavity for inspection. The patient is then asked to move the tonguesideways and up-and-down. The SLP looks for any sign of numbness ortingling in the face, cheeks or lips when the patient is following instructions.After that the SLP does a oral motor exam to test the strength of the musclesof the mouth. The SLPs hold a tongue depressor against the patients’ tongueand the patient is asked to push the tongue depressor outward, lift it up andpush it down. After the oral-motor exam the patient is asked to swallowwater sequentially until the glass is empty. The SLP places his/her hand ofthe patients throat to feel the swallow. After the swallow is done, the SLPinitiates a conversation with the patient to see if the voice is wet or gurgely.131.4. Diagnostic Evaluation of DysphagiaFigure 1.6: A modern videofluoroscopy machine used to perform swallowingassessment. Adapted from [46] with c licence via Wikimedia Commons.If there were any water was stuck at the throat, the patient would have had awet gurgely a voice. The clinician waits about a minute after the swallow toobserve if there is any cough or chocking that might be caused by potentialaspiration [44].The bedside assessment is effective in identifying disorders in the oralpreparatory and oral phase of swallowing. However, for evaluating pharyn-geal aspects of swallowing, bedside assessment is not very useful because it isdifficult to access if any bolus has entered the airway with a physical exam.Some patients suffering from neurological impairments may not cough orproduce gargley voice even if bolus has entered the airway during a bedsideexam[45]. Patients who show signs of dysphagia upon a bedside swallowingassessment are referred for comprehensive instrumental swallowing assess-ment.141.4. Diagnostic Evaluation of DysphagiaFigure 1.7: Lateral view of a health individual swallowing during an modifiedbarium swallow (MBS) exam. (a) Oral preparatory phase, (b) oral phase, (c)pharyngeal phase and (d) esophageal phase. Created from normal swallowMBS video used in MBSImPTM c© training, with permission [17].1.4.2 Modified Barium Swallow Study (MBSS)The Modified Barium Swallow Study (MBSS), also known as video-fluoroscopic(VF) swallow study, is a common, standard procedure and has been histori-cally reagarded as the gold standard for evaluating the swallowing mechanism[47; 48]. This test is often considered instrument of choice because it providesvisualization of the bolus flow in relation to structural movement throughoutupper aerodigestive tract. Such visualizations allow clinicians to robustlyand consistently identify the nature and severity of swallowing impairment,and the presence and timing of aspiration.The MBSS is usually conducted by a radiologist and a speech languagepathologist (SLP). During a MBSS, the patient is seated inside a VF ma-151.4. Diagnostic Evaluation of Dysphagiachine and presented with a bolus enhanced with barium to make the bolusvisible in the fluoroscope (Figure 1.6). This swallowing assessment capturesa sequences of videofluoroscopic images of bolus travelling through the oralcavity, pharynx and esophagus in real time. The patient swallow is initiallyviewed from lateral plane to evaluate the bolus transport during the oraland pharyngeal phase (Figure 1.7). This lateral view allows the clinician tomeasure the oral and pharyngeal transit times and presence of aspiration.The swallow can also be viewed from anterior-posterior plane to evaluateesophageal clearance. Boluses with different volume and consistencies arepresented and clinical impressions of the presence and degree of swallow-ing impairment are obtained from the VF images [49; 50]. The cliniciansalso make judgment regarding the coordination and timing of the swallowingevents based on the qualitative observation from the VF [51]. Furthermore,longitudinal MBSS is conducted to monitor any changes in swallowing func-tion over time to track the progression of a disease or condition.The disadvantage of MBSS is that it exposes the patient to a small amountof radiation. Although the dose of radiation is standard and may not poseany immediate threat, the clinician are cautious not to prolong the patientsexposure to radiation. Moreover, VF is not readily available everywhere andmay not be suitable for all patients e.g. patients with poor sitting posture,pregnant women, infants. Nevertheless, VF remains the standard swallowingassessment procedure and effective technique to evaluate level of aspiration.1.4.3 Endoscopic Evaluation of SwallowingThe use of flexible laryngoscopy to evaluate oropharyngeal dysphagia wasfirst published in 1988 [53]. This procedure, commonly known as FiberopticEndoscopic Evaluation of Swallowing (FEES), was developed as alternativeprocedure where it was not feasible for the patient to undergo a MBSS i.e.patients in ICU, care facility. FEES is one reliable method to assess thestructural and functional status of the oropharynx and larynx, during theswallowing process. A thin, flexible tube, called a laryngoscope, is used thathas a small camera on the end and is placed transnasally along the floor ofthe nose and advanced until end of the scope reaches the base of uvula. Thescope is sometimes advanced to the tip of the epiglottis to get a better viewof the pyriform sinuses and endolarynx, which is sometimes called protec-tive“cup” that guides material around the airway until the swallow occurs.A FEES assessment provides a good view of the larynx and partial view161.4. Diagnostic Evaluation of DysphagiaFigure 1.8: An illustration showing the flexible endoscopic tube placementbehind the soft palate to view the pharyngeal movement during a swallowingmotion. Adapted from [52] with c licence via Wikimedia Commons.of nasopharynx, oropharynx and hypopharynx (see Figure 1.9). Hence, theoral phase of swallow can be partially evaluated by FEES. The laryngoscopecannot go beyond the UES opening into the esophagus, the esophageal phaseof swallow cannot be accessed by the FEES examination.An endoscopic evaluation can be divided into two major parts: observa-tion and presentation of bolus. Once the scope is placed inside the nasalcavity the test begins. In the first phase the SLP assesses the anatomy of thepatient to determine the potential of the patient to execute a normal swallowwithout presenting any bolus. The clinician take notes regarding the surfaceanatomy of nasopharynx, oropharynx, and hypopharynx. Any alteration inthe anatomy from surgery, trauma can impair the normal swallowing physi-ology. The patient is referred to a otolaryngologist if there is a presence ofany foreign bodies and masses. The SLP also assesses sensory physiologyand may lightly touch the aryepiglottic fold or the tip of the epiglottis toinduce a cough. These observation and tests help the clinician to focus thesearch to look for any particular dysphagic patterns that may present.In the next part of the test, the patient is presented with a bolus to test171.4. Diagnostic Evaluation of DysphagiaFigure 1.9: The view of larynx captured during a FEES assessment. (a)View of the glottis and vocal cords as seen during an endoscopic procedure.Adapted from [54] with c licence via Wikimedia Commons. (b) A topview of the larynx: epiglottis, vocal cords, trachea, and cartilage are labeled.Adapted from [55] with c licence via Wikimedia Commons.the ability do a normal swallow. This is the major part of the test where theexaminer directly assesses the patient swallowing of food or liquid with theendoscope placed inside the nasal cavity. It is recommended to use a smallamount i.e. 5cc of a thin and/or thick liquid bolus and/or a bite of cracker[56]. With this test the SLP can evaluate and quantify bolus clearance,amount of aspiration and extent of airway closure. As an outcome of a endo-scopic assessment of swallowing, the SLP formulates an accurate impressionabout the nature of the problem and makes realistic recommendations.1.4.4 ManometryManometry is a test used to detect swallowing abnormalities after the bolushas entered into the pharynx and traveling towards stomach via esophagus.This test is used to measure pressures in the pharynx and/or esophagusduring a swallowing motion. The measured pressure quantifies strength andmuscle coordination of the pharynx and esophagus. MBS study can also beused simultaneously to observe indirect effects of pressure adequacy duringswallowing.During the manometry test, a small flexible tube is placed through thenose and into the pharynx and esophagus. Pressure sensors on the tube181.4. Diagnostic Evaluation of DysphagiaFigure 1.10: During manometry, a manometry catheter is passed through thenose, along the back of the throat, down the esophagus and into the stomachthrough the esophageal sphincter valves. This catheter records pressure forsubsequent anatomical regions (shown in right). Adapted from [57] with clicence via Wikimedia Commons.records pressure throughout the swallow. If pressure is not adequate at anylevel of the pharynx, food will remain at that level, rather than being pushedalong to the next level of the digestive tract. This test indicates how wellthe esophagus can perform peristalsis and allow the clinician to examinethe muscular valve connecting the esophagus with the stomach called LowerEsophageal Sphincter (LES). LES relaxes to allow food and liquid to enterthe stomach, and closes to prevent reverse flow of bolus moving out of thestomach and back up into the esophagus. Abnormalities with peristalsis andLES function may cause symptoms such as swallowing difficulty, heartburn,or chest pain [58]. Patients usually perceive this as food or fluid left over aftertheir swallow. Figure 1.10 shows a schematic of manometry done for a patientdiagnosed with achalasia. Achalasia is a condition in which the muscles ofthe lower part of the esophagus fail to relax, preventing food from passinginto the stomach causing esophageal dysphagia. Figure 1.10 also shows thataperistaltic contractions, increased intra-esophageal pressure, and failure of191.4. Diagnostic Evaluation of Dysphagiarelaxation of the LES.1.4.5 UltrasoundFigure 1.11: An ultrasound scan of oral complex during a swallow showingmidline sagittal scan of the tongue from the tip to dorsum. The ultrasoundtransducer is held beneath the patients chin. Adapted from [59] with clicence via Wikimedia Commons.Ultrasound is an imaging modality that uses high-frequency sound wavesto view the internal structures of the body. Unlike VF, no ionizing radia-tion exposure is associated with ultrasound imaging. Ultrasound is one ofthe most widely used imaging technologies found in medicine because it isportable, free of radiation risk, and relatively inexpensive when comparedwith other imaging modalities. Ultrasound has also been used to evaluateand access the swallowing process [60; 61]. Ultrasound is particularly usefulin swallowing assessment of infants because it allows the clinicians to evalu-ate infants suck/suckle type feeding on breast or bottle. US evaluation forinfants are preferred over other swallowing assessment techniques because itis non invasive and radiation free.201.5. Clinical Decision MakingNewer ultrasound units are now capable of identifying normal and patho-logical oropharyngeal tissues and visualizing the coordinated movements ofintegrated structures within the oropharynx [62]. Sagittal sonography of thetongue makes it possible to evaluate the tongue at rest and during swallow-ing, and oral preparatory and oral phase of swallow can be clearly imagedwith ultrasound. However, a limitation of ultrasound technique can is thatsound wave cannot pass through air or bone, rather it will be completelyreflected [63]. So ultrasound technique cannot visualize the trachea as it isan air-filled space. Although, ultrasound can detect pooling of secretion andresidue in the valleculae [64], it is not able to detect penetration or aspira-tion of contents into the trachea. Another limitation is that the hard palatecannot be visualized as it is a bony structure, making it difficult to evaluateglossopalatal function and adequacy. These factors limits the use of ultra-sound in assessment of pharyngeal phase of swallow [65]. Ultrasound is alsouseful for infants/children where repetition of the swallowing test may be es-sential and ultrasound scanning of the swallowing sequence can be repeatedwithout any comparable radiation risk [66].1.5 Clinical Decision Making: WhichSwallowing Assessment is Indicated?Different swallowing assessment provides different information about theswallowing physiology. Not one particular test can be used to evaluate swal-lowing abnormality rather different tests depending on the clinicians rec-ommendation should be preformed to get a comprehensive picture of theabnormality for planning the treatment. For example, though MBS study isregarded as the gold standard [47; 48], repeated testing is not recommendeddue to the risk of redundant radiation exposure. On the other hand somestudies have shown that FEES have a high level of agreement when one istrying to detect aspiration [67; 68; 69; 70], but this high level of agreementdoes not make the other test redundant. Furthermore, as discussed in section1.4.3, endoscopic evaluation partially evaluate the oral preparatory and oralphase of swallow, and cannot evaluate esophageal phase of swallow. Whatit means is the indication towards a assessment technique are based on thenature of the suspected problem and the clinician should have all the proce-dures available so that the appropriate one can be used with each patient.211.6. Standardized Evaluation of Dysphagia DiagnosisSome indication can also be logistic i.e. a FEES procedure requires a spe-cialized equipment and skilled operator which may not be available at allsettings.1.6 Standardized Evaluation of DysphagiaDiagnosisStandardized health care practices have been shown to improve clinical out-come by reducing ambiguous reporting and interpretation of test results [71].Lack of such standardization in measurement methods produce ambiguousresults when comparing results across heath care settings. Swallowing assess-ment using MBSS is a test of physiologic swallowing function and an indirectmeasure of sensory and motor functionalities that differentiates between anormal and impaired swallow. Variability in the MBSS result interpretationcan result in inauspicious patient outcome and impact the overall health andwell being of the patient.Many swallowing screening and instrumental evaluation techniques havebeen developed for diagnosis of dysphagia (discussed in Section 1.4). Thesetools evaluate pressure, range, strength of structural movement, airway pro-tection, sensation, bolus clearance and efficiency, and bolus flow patterns thatcharacterizes a normal and impaired swallow [72]. During a swallow screen-ing, the examiner looks for sings that could indicate dysphagia in a quick,efficient and safe manner. However, these screening tests does not give ac-tual physiological information. Meanwhile, a swallowing assessment is morecomplex and sometime involve invasive measurement for providing providingquantitative and qualitative data about swallowing physiology. There aremany studies in the literature conducted reviews to determine the effective-ness and feasibility of different screening and instrumental evaluation testsdone to diagnose dysphagia [73; 74; 75]. Some research groups have com-pared similar findings from two studies and concluded that both have highlevel of agreement [67; 68; 69; 70]. So it is a subjective decision to choose thecorrect method of evaluation which depends on what the clinician is lookingfor and a single test cannot possibly provide the best assessment for everypatient in every condition. Nonetheless, MBSS are preferred method by mostof the clinicians to diagnose dysphagia and also used to monitor any changesin the swallowing function over time to tract the effectiveness of swallowing221.6. Standardized Evaluation of Dysphagia Diagnosistreatment [72].1.6.1 Modified Barium Swallow Impairment ProfileMartin-Harris et al. [17] established a protocol for standardizing MBSS calledModified Barium Swallow Impairment Profile (MBSImP TM c©) which wasdeveloped during a five year study on over 300 dysphagic patients. This pro-tocol establishes universally accepted terminology for describing swallowingmotion and converts the clinical qualitative information into a quantifiablemetric to diagnose swallowing impairment. MBSImPTM c© standardizes thevolume, consistency and texture of the bolus administered during a vide-ofluoroscopic evaluation. It includes 17 physiologic swallowing componentsgrouped across three functional domains of swallowing (listed in Table 1.2)starting from lip closure to esophageal clearance. Each component is scoredindividually which uniquely contributes to judgement of overall swallowingimpairment [17; 76]. A component is scored from 0 to 3,4,5 where a score of“0” for a particular component would mean normal functionality and highernumber indicating a worse impairment (see Figure 1.12 for an example). De-tailed component list, corresponding scores and score definitions defined inMBSImPTM c© are given in Appendix A.Figure 1.12: Operational definition used to score “Component 6 - Initiation ofPharyngeal Swallow”. A score of 6-0 means bolus head (inside yellow circle)at posterior angle of ramus (first hyoid excursion), 6-1 means bolus head invalleculae, 6-2 means bolus head at posterior laryngeal surface of epiglottisand 6-3 means bolus head in pyriforms. A score of 6-4 (not shown in thefigure) would mean no visible initiation at any location. Image sequence iscreated from MBSImPTM c© online training videos with permission.231.6. Standardized Evaluation of Dysphagia DiagnosisOral impairment domain(1) Lip closure (Lip C)(2) Hold position/tongue control (HP)(3) Bolus preparation/mastication (BP)(4) Bolus transport/lingual motion (BT)(5) Oral residue (OR)(6) Initiation of the pharyngeal swallow (IPS)Pharyngeal impairment domain(7) Soft palate elevation (SPE)(8) Laryngeal elevation (LE)(9) Anterior hyoid motion (HM)(10) Epiglottic movement (EM)(11) Laryngeal closure (LC)(12) Pharyngeal stripping wave (PSW)(13) Pharyngeal contraction (PC)(14) PES opening (PESO)(15) Tongue base retraction (TBE)(16) Pharyngeal residue (PR)Esophageal impairment domain(17) Esophageal clearance in the upright position (EC)Table 1.2: List of physiologic swallowing components and correspondingacronyms used in MBSImPTM c© swallowing assessment.1.6.2 Reliability training and testingInter-and intra rater reliability of MBSImPTM c© protocol is required to be80% for blinded scores of modified barium swallow videos. In the training,if the concordance fell below 80%, the scores are reviewed and the trainingcontinued until the minimum of 80% concordance was achieved. The lowerinter- and intra-rater variability of the training is, the better the clinicians canaccurately and consistently score based on comparison to the standard. Spe-cialized training including accuracy and reliability measurement is requiredto train clinicians to score each swallow in a standardized manner. To ad-minister such training and ensure that the required reliability is maintainedwhile scoring following MBSImPTM c© protocol, an online training platformhas been developed [76]. The online training platform include standardized241.7. Biomechanical Model in Standardized Dysphagia Trainingtraining in swallowing physiology, skill development and reliability testing.The raining includes detailed animation derived from actual patient MBSSdata for all 72 possible different scores. After completing the learning zonethe trainee takes the reliability test.1.7 Biomechanical Model in StandardizedDysphagia TrainingThe oropharyngeal anatomy is complex and composed of rigid structures in-cluding the cranium, jaw, and hyoid bone, highly deformable muscle activatedtissues such as the tongue, soft palate, and pharynx, and larynx, an intri-cate arrangement of many muscles. These muscles and structures preformcomplicated and coordinated action like swallowing, mastication, chewingand speech. To understand such complicated motions and coordination todiagnose any dysfunction requires a sophisticated model that accounts formechanics as well as dynamics of the oropharyngeal complex. Understand-ing of the swallowing motion requires 3D visualization of the swallowingmotions which requires a 3D model representing the anatomical structures.So there is a need for a model that can provide 3D perspective visualizationof the swallowing motion that can help understand the complex dynamicsand help identify salient features of normal and impaired swallow. Further-more, swallowing dynamics changes with the change in viscosity of the bolusthat is being swallowed. So the model need to be able to simulate differ-ent consistencies of the bolus in order interactively learn how various bolusconsistencies effect the swallowing dynamics.1.8 ContributionsThe contributions of this Thesis include creating a 3D biomechanical modelof the oropharyngeal complex using realistic geometries and accurate timingof swallowing events (Chapter 3), simulating a stable bolus driven by modelkinematics (Chapter 4) and extension of existing training material for stan-dardized diagnosis of dysphagia (Chapter 5). The primary contributions ofthis dissertation are listed and summarized below:251.8. ContributionsModeling of the oropharyngeal swallowing motioni. Built realistic 3D geometries of oropharyngeal complex. Webuilt realistic full 3D geometries of the oropharyngeal complex consist-ing of tongue, jaw, hard palate, soft palate, hyoid, pharynx, thyroid,cricoid, epiglottis and arytenoid from a animation model used in clini-cian training.ii. Extracted accurate timing of swallowing events. We extractedaccurate timing of swallowing events from video animations used in stan-dardized clinician training for dysphagia diagnosis.iii. Created a coupled oropharyngeal swallowing model. We gener-ated rigid body and finite elements models from the surface geometriesextracted earlier to create a coupled biomechanical model of the oropha-ryngeal complex.iv. Driving the model kinematically to emulate swallowing motion.We used the kinematics derived earlier to drive our model usinng theaccurate timing of swallowing events.Fluid simulation to emulate bolus transporti. Incorporated the deformation of upper airway during swallowmotion. We used a airway-skin mesh for incorporating the deformationof the upper airway during a swallow motion.ii. Simulated stable bolus driven by model kinematics. We simu-lated a stable fluid bolus in the airway-skin driven by model dynamicsto emulate oropharyngeal swallowing motion in our model.iii. Validated simulation results with VF and animation video ofnormal swallow We demonstrated that our model can simulate a bolusin a manner consistent with the VF and animated illustrations used instandardized clinician training for dysphagia diagnosis.iv. Simulated standardized bolus consistencies. We also demonstratedthat our model can simulate boluses with different viscosity by simulat-ing standardized bolus consistencies used in clinical practice of dysphagiadiagnosis.261.9. Thesis OutlineEnhancing standardized training materials fordysphagia diagnosisi. Extension of training materials for most difficult physiologicswallowing component. We interviewed the SLPs to find out if theadditional 3D perspective visualization and flexibility of simulating fluidboluses with different consistencies can add value to their understandingof swallowing dynamics. With their feedback, we created 3D extensionof existing training materials for standardized dysphagia diagnosis.ii. Conducted a user study involving SLPs. The goal of the study wasto investigate if additional 3D visualization adds value in standardizedtraining. We built an interface similar to the existing MBSImPTM c© on-line training and added the training material afforded by our 3D model.iii. Concluded that our model can provide useful augmentation forstandardized dysphagia training. All the participating cliniciansrecommended with a high agreement score that the extension providedby our 3D model could be a useful addition for standardized dysphagiatraining.1.9 Thesis OutlineThe rest of the Thesis is organized as the following. First, an overview ofthe existing research in the relevant field is presented in Chapter 2. The full3D oropharyngeal model development and driving the model with accuratetiming of swallowing event is discussed in Chapter 3. Chapter 4 discussesabout the development of the computational fluid simulation techniques usedto simulate a bolus in the model and presents some qualitative and quantita-tive observation. Chapter 5 presents a user study conducted to evaluate thepotential of extending standardized training materials for dysphagia diagno-sis with our 3D model. Finally, the Thesis is concluded in Chapter 6 with asummary and some possible future directions.27Chapter 2Related WorkThe literature is dense with different models and simulation techniques aimedto emulate the oropharyngeal swallowing motion. Such models are intendedto be used in different application starting from dysphagia research to mod-eling different food texture. In this chapter, the literature will be reviewedto find out about different approaches taken for simulating a swallow motionand discuss about the remaining challenges that still need to be addressed inorder to develop a oropharyngeal swallowing for clinical training of dyspha-gia.2.1 Resources Used for Clinical Education& TrainingPhysiological models that attempt to reproduce living anatomy or physiologyare clinically called as “Simulator”. ‘Simulation” refers the application ofsuch simulator to perform a particular task for education or training purpose.Simulation in some form has probably been used as a teaching strategy ineducation and training processes for clinicians in all domains for a long time.It is often said that the first time simulation was done when the first doctortried to teach the first medical student how to perform a procedure properly.These simulators have a strong positive impact on health-care, as they lowerthe risk to patients of training and by providing a method of learning aboutcare processes.Illustrations of anatomic structures and animation of physiological pro-cess are also used for clinical training and they help explain a physiologicalprocess. Research is going on for developing more realistic, efficient, cheapand high-fidelity models to enhance such training materials. The followingsections describe some clinical training models relevant to the scope of thisThesis.282.2. Mannequin Simulator ModelsFigure 2.1: Two of the most recognised mannequin models. (a) Resusci-Anneand its later version are used for training CPR. Image adapted from [78].(b) Harvey and its upgraded versions are used to simulate cardiopulmonarypatient used mainly by medical students and cardiologists. Images (a) and(b) are adapted respectively from [78] and [79] c licence via WikimediaCommons.2.2 Mannequin Simulator ModelsHumanoid models for clinical education or training are often called man-nequins simulators. Mannequin simulators have been developed over theyears to be used for education, training and research purposes [77]. Theapplication of such mannequin models are in cardiopulmonary resuscitation,cardiology skills, anaesthesia clinical skills, fiberoptic endoscopic evaluationof swallowing, crisis management and many others. Virtual patient modelsare also used for training clinicians about a specific procedure and the de-velopment of computational power has enabled higher fidelity, more realisticsimulator of virtual patients. Both mannequin and virtual patient modelsare sometimes commercially available.The mannequin simulators were invented due to almost independent de-velopments that led to differences in technical approach in the early years.The first commercially available simulator model was “Resusci-Anne” (seeFigure 2.1(a)) which was developed specifically for practicing a cardiopul-monary resuscitation (CPR) in early 1960s [80; 77]. Later on, higher fidelitymodels were developed in the 1990s with more anatomically correct airway292.2. Mannequin Simulator Modelsand simulator called SimMan [80]. Around the time when Resusci-Anne wascreated, Harvey [81] another mannequin simulator was created. “Harvey”was a full sized mannequin that simulates cardiac conditions (see Figure2.1(b)). It was able to display various physical findings which included bloodpressure, bilateral jugular venous pulse wave forms, arterial pulses and soon. Such historical milestones contributed significantly in the developmentof realistic mannequin simulators.After the initial development of the mannequin simulators marked by“Resusci-Anne” and “Harvey” in the 1960s, simulation in health-care edu-cation and training started to gain acceptance. Mannequin simulators werefirst used for for teaching of swallowing examination in 1987 [82], for fiber-endoscopic training. For evaluating dysphagia fiberoptic endoscopic eval-uation of swallowing (FEES) is becoming popular but one of the obstacleis to equip trainees with the transnasal endoscopy skills needed to performthe procedure. To help the trainees by providing necessary tools for learn-ing, there are some head and neck mannequin models. However, trainingstudents with realistic simulation is costly. In addition to the purchase andmaintenance costs of the endoscopy equipment, mannequins simulators rangefrom $40, 000 for a basic mannequin to well over $95, 000 for a sophisticatedmannequin [83]. Furthermore, it is also reported in [84] that there are no orlittle difference in pass time on volunteers between clinicians trained usingmannequin simulators and clinicians trained on the non-lifelike tools, whereboth of the groups were faster and more confident when performing secondendoscopy on a volunteer. So research is more focused on cost effective mea-sures than developing more realistic and life-like simulator, when it comes tocreating models for oropharyngeal complex.There has been recent interest in the development of robotic devices forsimulating human swallowing motion. Such models are mainly intended formodeling “safe food” for patient suffering from dysphagia. One of the ther-apies to improve swallowing efficiency is to recommend dietary modification[10]. For example, thicker boluses are easier to swallow than thin boluses forsome dysphagia patients. So the SLPs suggest such modified diet to help thepatients stay healthy and also improve the swallowing motion.302.3. Animation ModelsFigure 2.2: MBSImPTM c© online training platform includes both videoflu-oroscopic data and animated illustrations of the full range of standardizedswallowing impairments. Image reproduced from MBSImPTM c© online train-ing [76], with permission.2.3 Animation ModelsDevelopment in the animation and rendering technology has enabled ani-mated illustration of physiologic procedures to be used in clinical training.The earliest mention of computer based animation for oropharyngeal com-plex can be found in [85] where an animation routine was used to reconstructoropharyngeal swallowing motion from biplane VF and dynamic CT images.This allowed somewhat more elaborate visualization of the mechanics of swal-lowing motion which was not possible with concurrent imaging technologies.Scholten et al [86] developed a multimedia program for teaching studentsabout the swallowing dynamics which they called the “The Dynamic Swal-low”. It was a combination of different media including animations, videoimages, text, diagrams, and voice- overs and allowed some sort of interactiveaccess to resources which was more effective than text-based material.MBSImPTM c© protocol [17] uses animated videos side by side with VF312.3. Animation ModelsFigure 2.3: (a) Occuals rift, a virtual reality head-mounted display. Thetop picture shows the front view, and the bottom one the rear view and thecontrol box [87]. (b) Application of such VR tools for assessment of studentsfor clinical skills using 3DiTeams learner [88].of patient swallow to help trainee clinicians understand the salient featuresof normal and different type of standardized dysphagic swallow (see Figure2.2). This has been proven effective as the trainee clinicians needed to passa reliability test with at least 80% or more reliability after the training wasdone. However, again, the additional visualizations provided with the VFare animated not generated using physics based simulation. So there is alack of realism in animated motions.Interactive models are an extension of the animated models and are af-forded by technological advances in computational efficiency, image renderingand interface technology. Several limitations in developing mannequin modelare discussed in the previous sections and it was suggested that training onstandardized patients or volunteers is much more effective than training on asimulator. A standardized patient is a healthy person who is trained to act asreal patient by accurately reproducing a history, physical and/or emotionalmedical scenario. However, there lies some obvious limitations concerningresources needed to have access to volunteers or standardized patients. Thisled to the development of VR simulation technology which has the following322.3. Animation ModelsFigure 2.4: Flexible endoscopy training with the Simbionix GI MentorTM[91].advantages over mannequin model and volunteer models.1. Standardized patients and sophisticated volunteers are expensive. VRcan help reduce the cost significantly and allow for real life scenar-ios with highly interactive, artificially intelligent and natural languagecapable virtual human agents.2. There are significant concurrent research going on for creating virtualreality headset like Oculus Rift [89] or Sony’s Project Morpheus [90].Such advance in VR technology is helping to create more realistic andinteractive video technology that can be useful for training cliniciansand medical students (see Figure 2.3).3. Standardized patients and volunteers can characterize a limited diver-sity conditions and demographics. This is limited by the availability ofhuman actors and their skills. This is even a greater problem when theactor needs to be a child, adolescent, elder, or when simulating a com-plex symptom presentation. Virtual patients can be tuned to mimicany demographic that may be necessary for the simulation.Virtual reality devices give the trainee clinician more interactive learn-ing experience than animated illustrations. However, all the illustration are332.4. Biomechanical Simulation Modelsanimated not simulated,the interaction with the model was limited by pre-defined set of activities and might not be suitable for simulating real clinicalconditions. Computer models for endoscopic evaluation was first developedin 1980s [92]. Innovation in computational power helped computer models tobe more effective and interactive. For an endoscopic training simulator (seeFigure 2.4), endoscopic images from a real patients are stored. When thedummy endoscope in inserted in the model, the simulator records the move-ments of the endoscope. Corresponding images from the storage is displayedin response to the endoscopy being performed in real time. As technicaladvances continue, this capability is expected to have a significant impacton how clinical training is conducted in psychology and medicine. However,there is also a need for a cost analysis study to determine if such simulationcan actually reduce the time of training.2.4 Biomechanical Simulation ModelsThe traditional approach to biomechanical analysis involves recording obser-vation of human movement and applying statistics to describe relationshipswithin the observations. Such observational studies have generated a wealthof data regarding human motor physiology. The following steps are neededto for generating a biomechanical model of the oropharynx.2.4.1 Oropharyngeal geometriesThe quantitative analysis of the swallowing process may enable a systematicstudy of care foods that are safe and offer some degree of comfort to patientssuffering from swallowing disorders. Few researchers have tackled the nu-merical simulation of swallowing in any detail. Movements of the esophagusand tongue have been discussed in relation to bolus transport and modeledmathematically. However, any discussion of dysphagia using the results ofa numerical simulation must include the retroflexion of the epiglottis andlaryngeal elevation because disorders relating to these movements are closelyrelated to dysphagia. Thus the geometries of the oropharyngeal complexneed to be realistic.342.4. Biomechanical Simulation Models2.4.2 Simulation of bolusThe bolus motion is numerically solved in order to simulate swallowing mo-tion in a coupled biomechanical model. Usually, numerical solution strategystarts from observing a physical phenomenon and defining characteristicsthat are relevant to the investigation. Some governing equations, initial orboundary conditions are defined to form a simplified mathematical model.The equations are numerically solved by discretizing the domain and obtain-ing a discrete representation of of the governing equations. Finally a set ofordinary differential equations are derived and translated into a computercode. Numerical solution for hydrodynamic equations involves similar stepsand can broadly be divided into two major categories depending on the com-putational framework: (1) grid based methods and (2) meshfree methods.2.4.3 Grid Based MethodsThe grid based methods implement a computational model constructed bynodes through a predefined nodal connectivity. The accuracy of such numer-ical approximations significantly depend on mesh topology i.e. shape, size.To discretize the description of the fluid flow the partial differential equationsare approximated. The partial derivatives can be approximated in differentways such as finite difference, finite volume, or finite element schemes us-ing a computational mesh. This grid based techniques was first proposedby Harlow et al. [93] in 1965 called Marker and Cell (MAC) method. Thismethod consists of markers which defines the position of the fluid and gridcells to reference the physical variables of the fluid such as velocity, pressure,temperature. This grid can be visualized as a computational mesh. Thismesh can be fixed in space and cover the whole fluid domain, or it can befixed with the fluid and move with the flow. The former approach is is calledEulerian method or spatial description and the latter is called Lagrangianmethod or material description. Both of the approaches are widely practicedin numerical methods with comparative advantages and disadvantages. Se-lection of any one approach depends on the problem description, thus arebriefly discussed in the following sections.352.4. Biomechanical Simulation ModelsLagrangian GridThe Lagrangian method describes the physical governing equations using ma-terial description and is usually represented by finite element method (FEM)[94]. In this method a mesh is fixed/attached to the fluid during the compu-tation process. Due to such attachment, the mesh moves as the fluid flows.Since the mesh is fixed, the time history of all the field variables at a materialpoint can be easily tracked for a moving fluid. Using an irregular mesh, aLagrangian description can easily adapt to irregular and complex geometriesand describes free or moving boundaries and material interfaces. Further-more, there is no convective term in the related partial differential equation,whereas an Eulerian approach requires an computationally expensive con-vective term. Due to these features, such material description is useful insolving problems where the deformation is not large. However, if the motionof the fluid becomes geometrically complex, the mesh undergoes severe defor-mation; the underlying numerical methods become too unstable to computea solution.Eulerian GridOn the other hand, an Eulerian approach describes the physical governingequations spatially which is typically represented by finite difference method(FDM)[95]. The computational mesh is fixed in space unlike the Lagrangianapproach where the mesh is fixed with the fluid. As a result, the mesh pointsand cells remain spatially fixed during the computation while the fluid flowsacross the mesh. Physical quantities like velocity, energy, etc of the flow arecalculated by simulating flux of mass, momentum and energy across meshcell boundaries. As the Eulerian mesh is fixed in space large deformationsin the fluid flow do not cause numerical problems like the Lagrangian grid-based methods. Hence, Eulerian methods are popular in computational fluiddynamics where the flow of the fluid is significant. However, it is difficultto track the time history of a field variable at a fixed point on the material.For complex geometries a complicated mesh generation process in requiredto convert the irregular geometry into computational domain. There alsoexists a trade off between the computational efficiency and resolution of thecomputational mesh as the Eulerian method requires a mesh over the wholecomputational domain. The coarser the mesh is the more computationallyefficient the method becomes sacrificing accuracy and vice versa. Table 2.1362.4. Biomechanical Simulation Modelssummarizes comparative features Lagrangian and Eulerian.Grid (L) Fixed with the material.(E) Fixed in space.Tracking (L) Movement of any point on materials.(E) Mass, momentum and energy flux across gridnodes and mesh cell boundary.Time history (L) Easy to obtain at a point attached on the mate-rial.(E) Difficult to obtain at a point attached on thematerial.Moving boundary (L) Easy to track.(E) Difficult to track.Complex geometry (L) Easy to model.(E) Difficult to model with good accuracy.Large deformation (L) Difficult to handle.(E) Easy to handle.Table 2.1: Comparative features of Lagrangian(L) and Eulerian(E) gridbased methods for solving hydrodynamic equations.Combined Lagrangian and Eulerian gridsBoth Lagrangian and Eulerian approach have comparative advantages anddisadvantages. In literature, two hybrid methods are found to combine thesetwo to benefit from the advantages and minimize limitations. They are theCoupled Eulerian Lagrangian (CEL) [96] and Arbitrary Lagrange Eulerian(ALE) [97; 98]. The CEL approach is often undertaken when there are twodifferent types of materials with different level of deformation. The commonpractice is to discretize the one with little or no deformation, usually solidmaterial, in a Lagrangian frame; and fluids or material have significant defor-mations using Eulerian frame. Both Lagrangian and Eulerian descriptionsare setup in separate regions in the problem space and does not overlap.These independent regions interact with each other and exchange computa-tional information between two set of grids. In Arbitrary Lagrange Eulerian(ALE) approach rezoning techniques for Lagrangian meshes are employed.Such rezoning aims to minimize mesh distortion when the Lagrangian mesh372.4. Biomechanical Simulation Modelsis moved independently. In ALE the Lagrangian motion is computed at thebeginning of each time step and is followed by rezoning step. During thisrezoning step it is decided if the mesh is going to be rezoned to the origi-nal shape (Eulerian description), not rezoned (pure Lagrangian description)or some optimal shape (somewhere in between Lagrangian and Eulerian de-scription). Many commercial solver i.e. LS-DYNA[99], MSC/Dytran[100]uses such hybrid approaches and has received much research interest. How-ever, even with these hybrid approaches a severely distorted mesh causeserrors in the numerical simulations [97; 98] and often causes termination ofthe computation process.Limitations of the grid based methodsGrid based methods have been widely used in various areas of computationalfluid dynamics(CFD) and computational solid mechanics (CSM). However,there are some inherent difficulties undertaking such grid based approacheswhich limit their application in many fluid simulation problem. For ex-ample, if taking a Eulerian approach like FDM, it is always challenging toconstruct a regular grid to accommodate complex geometry. This requirescomplex mathematical transformations which can be more expensive thanthe problem itself. Further, such approach has problems with tracking ma-terial properties, dealing with deformable boundaries and free surfaces asshown in Table 2.1. Meanwhile, Lagrangian FEM methods requires rezoningwhich is expensive and introduces additional inaccuracy to the solution. Thelimitations of grid based methods are more evident when considering largedeformations, large inhomogeneities, moving material interfaces, deformableboundaries and free surfaces common in hydrodynamic phenomenon as explo-sion and high velocity impact (HVI). These limitations initiated the researchon meshfree methods that is required to solve these problems.2.4.4 Meshfree methodsThe basic idea behind the meshfree methods is to discretize the continuumthrough a set of points which are not connected by a computational mesh.382.4. Biomechanical Simulation ModelsSmoothed particle Hydrodynamics (SPH)Smoothed Particle Hydrodynamics (SPH) is one of the most popular mesh-free representation methods for simulating free surface flows. SPH was firstproposed independently by Gingold and Monaghan [101] and Lucy [102] inlate seventies. It was developed to study astrophysical phenomena as thesheer scale of the astrophysical events deem experimental setups in a labo-ratory setting impossible for most of the cases. SPH has been used to studya range of astronomical events such as galaxy formation, star formation, su-pernovas, solar system formation and so on. SPH became popular in manyother ares such as computational fluid simulation, solid mechanics and hasvarious application in coastal engineering for modeling dam brake behavioursand plunging waves, virtual water simulation for video games or motion pic-ture, virtual surgery, multifluid simulation and becoming increasingly popu-lar [103]. As SPH defines a set of moving interpolation points which movesto represent the fluid flow; in a sense SPH is a Lagrangian motion even if thepoints are never linked via a computational mesh. The interpolation pointsassume values of dynamic fluid variables which are expressed as an integralinterpolant using a smoothing kernel. The integral is approximated by asummation over the interpolation points. A detailed description about SPHfundamentals and formulation is given in Chapter B.There are several advantages of the SPH method over the traditional gridbased numerical methods [104]. By properly deploying SPH particles at theinitial positions; the free surfaces, material interfaces and moving boundariescan be traced regardless of the complexity of the fluid movement which is verychallenging to incorporate using Eulerian methods. Free surface representa-tion of fluids allow a straightforward handling of very large deformations.This is because the computational mesh representing the fluid domain is notdefined at the beginning rather the connectivity among the points is gen-erated as a part of the computation at each time step. Furthermore, theaccuracy can be controlled with flexibility and by adding more points (i.e.high resolution simulation) in the area of interest can give more accurateresults when refinement is needed. Such characteristics makes SPH suitablefor bio- and nano- engineering at micro and nano scale, and astrophysics atastronomic scale.392.5. SummaryUpper airway modelA pure Lagrangian mesh representing a liquid bolus can be used to simu-late the movement of bolus during a swallow. However, such meshing re-quires computationally expensive operation while simulating bolus mergingand splitting bolus, requiring frequent remeshing. An Eulerian representationlike Sonomura et al. [105] would require special handling of the irregular mov-ing boundaries. Fluid simulation of swallowing is complex as it is dependedon the irregular boundary motions provided by the anatomical structuresof the oropharyngeal complex like tongue, soft palate and pharyngeal wall,which are subject to individual variation. In this thesis, SPH is used to sim-ulate the bolus inside the oral cavity to emulate swallowing motion to takeadvantage of its mesh free formulation surface free representation. SPH hasalso been used to simulated both compressible[106] flow and incompressibleflow[107; 108].2.5 SummaryTo summarize, human swallowing motion is very complex and hard to visu-alize even with the most state-of-the-art imaging technology. To comprehendthe complex swallowing motion and investigate different swallowing impair-ment, realistic illustrations such as animations, 3D virtual models contributessignificantly and helps clinicians learning. These illustration are primarilycreated from observation of the structural movements from medical images.However, the motion portrayed by the animators are plausible but sometimesnot realistic as the anatomical structures are made to follow a certain motionthat do not support the underlying kinematics causing it. As a result, thislimits the interaction with the model limited and predefined where as realpatient cases are more complex and specific.Three areas of investigation were identified for generating biomechanicalmodel of the oropharyngeal complex for simulating a swallowing motion.First Getting realistic geometries of the oropharyngeal complex and accu-rate timing of swallowing events is challenging even with state-of-the-art structural and functional medical imaging data.Second Stable fluid simulation techniques for incorporating deformable bound-ary condition are required to mimic the bolus transport of a swallowingmotion in a biomechanical model.402.5. SummaryThird The potential of a simulation based training model for being acceptedat clinical practice largely depends on its effectiveness in clinical settingand its potential of improving the patient outcome.In the following three chapters, the contribution of this Thesis towardsthese open research problems are described.41Chapter 3Modeling of the OropharyngealComplexThis Chapter presents the process flow undertaken to create a 3D biome-chanical model of the oropharyngeal complex for swallowing simulation. Thefirst step towards getting the biomechanical model is building realistic ge-ometries of oropharyngeal structures. In our process flow, the generation of a3D biomechanical swallowing model involves several steps: building realisticgeometries, formulating a biomechanical model with the derived geometries,addling accurate timing of swallowing events into the model, driving themodel kinematically for a simulating a swallow. Each step of this processflow is explained in detail in the following sections of this chapter. Whilethe process flow was developed for the purpose of generating a biomechan-ical model from swallowing animations, it can also be applied for creatinggeneral-purpose biomechanical models from animated illustration for otherphysiological process.3.1 Building GeometriesModeling the highly geometrically complex oropharyngeal structures arechallenging because the oropharyngeal complex is made of soft tissues, mus-cles, bones and empty space in the airway. Such diversity makes it verydifficult for any particular imaging modality to comprehend different typeanatomical structures present in the oropharyngeal complex. Using realisticgeometries for a biomechanical simulation is important, especially when themodel is intended to be used for clinician training.The MBSImP c©TM training protocol include 72 videofluoroscopy seg-ments, each combined with animation videos to illustrate 17 physiologic swal-lowing components and scores of the MBSImP c©TM protocol. Appendix Alists the 17 physiologic swallowing components and consequent scoring of thecomponents following MBSImP c©TM. The animations were created using an423.1. Building Geometriesunderlying 2.5D geometric model (i.e. only a lateral side model is used) withshape changes specified over time for the animation. The geometries used tobuild the model were validated for clinical training of dysphagia by a team ofexpert clinicians. The animators consulted anatomy reference books, medi-cal images i.e. CT, VF, MRI from patents data to build the geometries ofthe oropharyngeal complex. The geometries and timing of the swallowingevents were iterated by the clinicians and animators until both were in anagreement that the animations exhibit the salient features of a swallowingmotion. Leveraging the animators imagination guided by expert clinicians,the MBSImP c©TM swallowing animations represent realistic geometries andvalidated timing of swallowing events. In this work, the animation modelused to create MBSImPTM c© animation videos is leveraged to build our 3Dbiomechanical model. The kinematics from the animations are propagated tothe biomechanical model to drive the model with clinically validated timingof swallowing events.3.1.1 Generating full 3D modelThe animation model(see Figure 3.1) for the was created in LightWave R© 3Dwhich is a registered trademark of NewTek Inc. It is a commercial softwarepackage extensively used to create visual effect, video games developmentand motion graphics. The surface geometries of the oropharyngeal complexwere artistically designed in LightWave Modeler R©. The model was designedin such a was that it would illustrate the oropharyngeal structures from midsagittal cut plane. The model was designed in this half cut way because theanimators wanted to render the videos from midsagittal cut plane to matchthe viewpoint of a usual MBS exam.After the surface geometries were created, the animators used LightWaveLayout R© to arrange the 3D model and define shape changes into the surfacegeometries in key-frames to animate a swallowing motion. This is to benoted that the kinematics are also added artistically to the model to matchthe input VF data. After the animators and clinicians were in agreement thethe kinematics exhibited by the model matched that of a real swallow, theanimator then manually drew some line in rostral-caudal axis and animatedsome sort bolus like structures to follow that predefined bolus path. Withthe key frame animation to deform/move the oropharangeal structures, thebolus is moved in sync with the kinematics of the model to match a normalswallow. After the model was rendered to create a normal swallow, video was433.1. Building GeometriesFigure 3.1: Animated swallowing model of the oropharyngeal built builtusing LightWave R© 3D.captured from mid sagittal cut plane to make the animated videos. Figure3.2 show different instances of key-frame animation for a normal swallowand corresponding frame in the swallowing video. For creating videos ofdifferent impaired swallow, these steps are repeated starting from the key-frame animation. It is to be noted that the surface geometries are createonce and different shape change controlled by key-frame animation enabledthe animators to animate different types of swallowing impairment and rendervideos to be used in the MBSImP c©TM training.The half cut surface geometries are extracted from the LightWave Modeler R©for building full 3D geometries. The animation model also included surfacerepresentation of human skeleton, eyeballs, lungs, hand and esophagus. How-ever, for the scope of the this Thesis, 3D surface geometries of tongue, hardand soft palate, teeth , arytenoid, hyoid, jaw, thyroid, cricoid, epiglottis andtrachea are built as the simulation of oropharyngeal swallow is of interest.443.1. Building GeometriesFigure 3.2: The top row shows instances of the animation model and thebottom row shows corresponding frame in the swallowing videos used inMBSImP c©TM standardized training.As mentioned earlier, the geometries are cut in half so there is a surfacerepresenting the mid sagittal cut plane. The first step for getting the 3Dmodel is manually removing the vertices and faces of the mid sagittal plane.After the the surface mesh was mirrored with respect to mid-sagittal axis.To make the model close, the vertices new faces were defined to connect thehalf object from the left to right. This is to be noted that no new verticeswere created or deleted in order to connect the two part rather only new faceswere defined. This is very important not to introduce new vertices into thesurface geometry. This is because in the future sections (Sections 3.2.1 and3.2.3), when sequential meshes are generated to extract kinematics from theanimation and use this kinematic information to drive the model, withoutone-to-one relation between the base mesh and sequential meshed the clini-cally validated kinematics could not be preserved. Figure 3.3 illiterates thesteps undertaken to generate the 3D surface geometry for the tongue. Thisprocess is repeated for all the selected anatomies to be exported.The hard palate, soft palate, uvula, pharyngeal wall and oropharynx isrepresented using one object in the animation model. However, such repre-sentation is problematic for building a biomechanical model because differentpart of this big object undergo different amount of deformation. For example,the hard palate is a bony structure exhibiting no deformation whereas thepharyngeal wall deforms significantly during a normal swallow motion. Sothe big object was divided into hard palate, soft palate, uvula and pharyngeal453.1. Building Geometrieswall depending on their anatomical location and level of deformations.Figure 3.3: Steps for generating full 3D geometries of oropharyngeal com-plex, illustrated for tongue model. (a) The half cut object extracted fromLightWave Modeler R©. (b) First the mid sagittal cut plane surface is manu-ally removed, afterwards (c) the object is mirrored and (d) the correspondingvertices are connected to create new faces and make the geometry bounded.3.1.2 Realistic morphometryAfter generating 3d surface model for the oropharyngeal complex the model isscaled to represent the dimension of average human oropharyngeal anatomy.Our model needs to be morphometrically realistic in order to produce mean-ingful clinical results. Having realistic shape and size enable a biomechanicalmodel to generate clinically meaningful result. To make the model mor-phometrically realistic, the tongue was selected to the anatomy of reference.This is because, the tongue is of major interest to phoneticians, linguists,speech language pathologists and orthodontists. The size of the tongue is463.1. Building Geometrieshighly variable between sex, age of the individual. Hopkin et al.[109] mea-sured the dimension of 32 neonatal tongues (16 male, 16 female) and 30 adulttongue (14 male and 16 female) postmortem. The findings regrading meandimension of neonatal and adult tongue from that study are reported in thefollowing table.Neonatal (N = 32) Adult (N = 30)Mean SD SE Mean SD SELength 39.91 4.25 0.76 79.83 7.57 1.40Breadth 25.43 2.58 0.46 51.90 4.24 0.78Thickness 8.76 0.98 0.17 16.13 2.51 0.46Table 3.1: Mean dimension in mm of neonatal and adult tongue reportedby [109]. SD and SE stands for Standard Deviation and Standard Error ofmean respectively.The above table shows that mean dimension of an adult tongue are aboutdouble of a that of a neonatal. As the aim of this Thesis is to generate a modelfor clinician training and offer a generic representation of the population,adult tongue dimension was considered to in order to scale the model. Amongthe three attributes of tongue dimension reported, the one with the leaststandard deviation should be selected. Tongue length was measured from thetip of the epiglottis to the apex of the tongue. The breadth was measuredat the widest part of the tongue and the thickness was measured at the freeedge of the tongue at its widest part. In Table 3.4, it can be seen thattongue thickness has the least standard deviation. However, the free edgeof the widest part of the tongue is not clearly defined and hard to measurein the model. The tongue length has higher standard deviation than tonguebreadth, so tongue breadth was selected to calculate the scaling factor formaking our model morphometrically realistic. It can be seen from Table3.4 that the mean tongue breadth for adults was found to be 51.9mm. Forcalculating the scaling factor, the widest part of the tongue model was scaledto have 51.9mm in breadth (Figure 3.4). The same scaling factor is then usedto scale the whole model. By doing so, we make our model morphometricallyrealistic.473.2. Building the biomechanical modelFigure 3.4: Average tongue breadth simulated in the model to match thedimensions reported in [109]3.2 Building the biomechanical modelThe MBSImP c©TM animations exhibit accurate timings of the swallowingevents which is validated for clinician training and this accurate timing ofswallowing events are incorporated in the biomechanical model. This sectionwill discuss about extracting the kinematics from the animation, building acoupled biomechanical model with the full surface geometries and derivingthe coupled biomechanical with the kinematics extracted from the animation.3.2.1 Extracting kinematics from animationTo build our biomechanical swallowing model, the full 3d geometries de-rived in the previous section (Section 3.1) are driven with the kinematicsextracted from the animation. This animation is introduced and renderedby LightWave Layout R© software to create videos for MBSImP c©TM swal-lowing animations. 109 frames starting from 0th the 108th time instant wasanimated to render the 3 second animations illustrating a normal swallow.The vertices position of the each surface mesh representing an object in theanimation scene was recorded from time instance 0 to 108 to capture theshape change of the object over the duration of the swallowing motion. Inother words, to capture accurate timing of swallowing events, a surface meshrepresenting the object was exported for 0 to 108 time instant producing109 surface meshes. As one surface meshes represent the same object at a483.2. Building the biomechanical modelFigure 3.5: Sequential meshes extracted from the animation model to capturethe accurate timing and deformation of the oropharyngeal structures duringa swallowing motion. This figure shows tongue movement during a swallowmotion captured by extracting the sequential meshes and it is repeated forall geometries of our model.particular time instant, the number of vertices and faces are same in all thesequential meshes, only the position of the vertices are changed. Figure 3.5show some of the sequential meshed generated for capturing the tongue mo-tion over time for a normal swallow. This process is repeated for to generatesequential meshed for tongue, hard and soft palate, teeth, arytenoid, hyoid,jaw, thyroid, cricoid, epiglottis. By exporting the sequential meshes for allrelevant objects in the animation scene, the clinically validates kinematics ofswallowing events are recorded.3.2.2 Coupled biomechanical modelOur biomechanical swallowing model of the oropharyngeal complex is builtand simulated in ArtiSynth (, a biomechanical simula-tion toolkit [110] that supports combined multi-body and finite element sim-ulation. The oropharyngeal complex is made of both soft and bony structureswho undergo different type of deformation/movement depending of its tis-sue property during a swallowing motion. For example, bony tissue like jawor hyoid bone undergo rigid transformations whereas the softer structures493.2. Building the biomechanical modelFigure 3.6: Different components of the 3D oropharyngeal biomechanicalswallowing the tongue, undergoes significant nonrigid transformation. So it is notpractical to simulate all the structures with rigid or nonrigid transformation.Figure 3.6 shows the simulated rigid bodies and FEMs in the model.Building rigid bodiesRigid bodies are implemented in ArtiSynth as dynamic components withsix dimensional position, orientation state with corresponding velocity statewith inertia. There is also a surface mesh associated with each rigid bodydescribing its topology. A rigid body has 6 Degrees Of Freedom (DOF) whichis described by its six dimensional position state and its pose is computedfrom its position and orientation with respect to world coordinates. Therigid bodies are created from the full 3D surface geometries of the bonyanatomical structures of the oropharyngeal complex built in Section 3.1. Afactory method used in ArtiSynth is used to create rigid bodies where thederived surface topologies are set as the surface mesh associated with the503.2. Building the biomechanical modelFigure 3.7: A surface is extruded along the normal of the triangular faces ofthe surface mesh creating FEM with 2 layer wedge element. The thicknesswhich is the distance between the layers are set to 0 to superimpose one layeron top of the other.rigid body. Uniform density is defined throughout from which mass andinertia properties are calculated.Building finite element modelsThe general approach for generating a FE model is by dividing the domaininto elements which are the cells or building blocks of the model. The ver-tices or corners of the elements are called nodes. The domain is a three-dimensional space occupied by the model and divided into small elementswhich accurately represent the model geometry. This elements allow to cre-ate local domains by approximating the system equations locally and thenodes act as the control points in the spatially discretized system. When theequation is solved, the solution is assumed to be smoothly interpolated acrossthe elements based on values determined at the nodes. This discretization ofthe differential system is converted into an algebraic system where it can belinearized and solved iteratively.Soft and deformable oropharyngeal structures are simulated as finite el-ement models (FEM). The first step for generating a FEM is to build the513.2. Building the biomechanical modelmesh topology i.e. node and element structure. The FEMs are generatedfrom the surface meshes extracted in Section 3.1. Our model is intended tobe driven by kinematics. So the FEMs are extruded to make a hollow FEMsto preserve the surface geometries. This makes it easier to drive the modelkinematically with less computational complexity. A thin layer of elementsare created along the normal direction by extruding the triangular faces ofthe surface mesh. As a result the triangular faces become two layer wedgeelements with a defined thickness i.e. distance between the layers (see Figure3.7). During extruding the triangular faces, the distance between the layersare set to zeros so two layers of FEM nodes are elements are superimposedon each other. As result the elements are inverted as the surface curvaturebecomes greater than total extruded thickness. This inversion of the elementis mitigated by setting the boundary condition explicitly. What it means isthe both position and velocity of the nodes will be controlled parametricallyby applying this fixed boundary condition and turning the dynamic parame-ter to “false”. After defining the surface topology, the material properties aredefined as linear and different rendering properties are added to the modelto make it look more realistic which is very important for it to be used invisualization application.3.2.3 Adding kinematicsAfter developing the biomechanical model with rigid bodies and FEMs, kine-matics are added to the model to mimic the oropharyngeal swallowing mo-tion. The model was built from the geometries representing the first timeinstant of a swallow motion that is 0th time instant. Sequential surfacetopologies are extracted form the animations in Section 3.2.1 to capture thetiming of swallowing events. The surface topologies were extracted for all thetime frames starting from 0th to 108th time instant. These sequential mesheshad same number of vertices and faces but different position in space cap-turing the deformation of the anatomical structure over time. The positionof the FEM nodes over time were updated from the corresponding sequen-tial mesh of that time instance to deform the FEM for emulating swallowingmotion. The one to one correspondence between the vertices of the sequen-tial meshes allowed to reliably propagate the swallowing kinematics into themodel. 109 sequential meshes are loaded to simulation a swallowing motionof 3.18 second. Figure 3.8 shows different instances of our model emulating anormal swallow motion which is superimposed on the animation video. This523.3. Results and AnalysisFigure 3.8: Kinematically driven biomechanical model superimposed on thenormal swallowing video.figure illustrate that kinematics of animations are propagated correctly intoour model.3.3 Results and AnalysisThe VF allows the clinicians to access the swallowing functionality of a pa-tients but for trainee clinicians, it is sometimes hard to comprehend theswallowing physiology from the VF. The animations helps the trainee clini-cians’ understanding of the swallowing physiology by illustrating the salientfeatures more clearly. However, the animations lack realism as the bolusmovements are animated not simulated. Furthermore, the animations aremade to capture the swallowing motion from lateral plane. Using our biome-chanical model, the motions can be simulated not animated. The advantageof simulating the motion not animating is that, it makes the model moreinteractive offering additional flexibility. With our model, the swallowingcan be viewed from different viewpoints in full 3D allowing perspective vi-sualization. This simulation can help incorporate the missing realism into533.4. Summarythe training, which in turn will bridge the gap between the VF and trainingmodel. Figure 3.9 shows the kinematically driven model superimposed onthe normal swallowing VF and animation.The generated sequential meshes extracted for the animation scene arehalf cut surface representation as the animation model was half cut. However,the model components are full 3D models representing the first time instanceof a swallow motion. So to update the position of the full 3D models withthe half cut sequential meshes, the one-to-one property of the sequentialmeshes is applied. As the sequential meshed represent the deformation ofthe same object overtime, the number of vertices in each sequential mesh issame. During the creation of full 3D geometries (Figure 3.3) no new verticeswere created only vertices of the mid sagittal cut plane were deleted. So atext file was generated for each time instance, starting for 0 to 108 only withthe Cartesian coordinates of the vertices from the sequential meshes. Thevertices of the left half of a object at the 0th time instance were updatedwith the positions of the vertices from the next time instance except forthe vertices that were deleted when the mid-sagittal cut place was removed.For the other half, the X coordinate was multiplied by −1 to to mirror theCartesian coordinates with respect to X axis. By doing so, the full object isupdated with the position of the next half cut sequential mesh. Loading onlythe text file with vertices not the sequential surface meshes make the processmore computationally efficient as it avoids unnecessary book keeping for eachloaded mesh in the ArtiSynth. This process is repeated for each time frameand for each FEM components for simulating a swallowing motion. By doingso, we propagate the kinematics and accurate timings of swallow events todrive our coupled biomechanical model.3.4 SummaryThe synergy of oropharyngeal swallowing dynamics is simulated in this Chap-ter. Full 3D oropharyngeal model was built from surface geometries extractedfrom the animation model. Rigid body models for bony structures and FEMsfor soft deformable structures of the oropharyngeal complex were generatedfrom the surface geometries. This coupled biomechanical model was kine-matically driven with accurate timing of swallowing events. This coupledmodel sets up the stage for further development where the model can becontrolled dynamically by activating certain muscles. ArtiSynth supports a543.4. SummaryFigure 3.9: Our kinematically driven biomechanical model superimposed onvideofluoroscopy (left column) and animation (right column) of a normalswallowing motion.553.4. Summarymuscle driven FE model where the muscle fibres embedded in the model canbe activated to follow a desired movement. It is possible to use the inversemodeling capability of ArtiSynth to estimate the virtual muscle activationfrom the kinematics of our implemented FE tongue for a normal and impairedswallow. This can give a deeper comprehension of the swallowing physiologyfrom a neurological perspective and aid dysphagia diagnosis.56Chapter 4Bolus SimulationIn Chapter 3, a biomechanical model of the oropharyngeal complex was de-veloped from realistic geometries and accurate timing of swallowing events.In this Chapter, a fluid bolus is simulated inside the kinematically drivenoropharyngeal model driven by the model kinematics. Section 4.1 describesthe unified geometric skinning technique used to couple the airway-skin meshwith our oropharyngeal model. The airway-skin mesh represents upper air-way which is deformed significantly by the movement of the oropharyngealstructures during a swallowing motion. Section 4.2 describes the SmoothedParticle Hydrodynamics (SPH) approach undertaken to simulate a fluid bo-lus inside the airway-skin mesh. The kinematics of the model geometry(including timings) characterized by changes in volume and shape, drive thecoupled airway-skin that in turn mobilizes the simulated bolus. Boluses withstandardized viscosity following National Dysphagia Diet Task Force speci-fication [111] is simulated in Section 4.3.2. Finally, Section 4.4 discusses theimplications of the bolus simulation and directions for future refinement.4.1 Deformable Airway ModelAirway is the empty space starting at the tip of the tongue through theoropharynx and ending at the starting of the esophagus. This internal skinsurface of the throat and mouth bounds the airway which covers hard andsoft palate, tongue, uvula and pharynx. This internal skin is similar toto the external skin of the body but rather than enclosing other anatomicalstructures, it is enclosed by the surrounding structures. We call it the airway-skin. The esophageal part of the airway is not included in our model becausewe only simulate the oropharyngeal phase of swallowing motion.The airway-skin describes the deformations of the empty space betweenthe anatomical structures that are deformed (soft tissues like tongue) andtranslated (bony structures like hyoid) during a swallow motion. As a result,the size, shape and surface area of the airway are greatly influenced by the574.1. Deformable Airway Modelmovement of the surrounding structures during the swallowing motion. Thusmodeling the airway-skin to incorporate the swallowing dynamics poses thefollowing challenges:• Our model is built as a mixture of rigid bodies and FEMs. So theairway-skin should be influenced by both type of dynamic components.• The airway-skin changes its shape, size and volume during a swallowingmotion. To incorporate the changes the airway-skin should deform ateach time step with the underlying model dynamics.• For simulating a swallow motion using the deforming airway-skin, abolus is simulated inside the airway-skin. So the airway-skin is requiredto have watertight geometry and be able to bound the bolus throughoutthe swallowing motion.• The bolus is transported from the tongue tip into the pharynx by theexplicit force applied to the bolus by the deforming anatomies of theoropharyngeal complex. Thus the airway-skin should allow forces tobe transmitted to the bolus, generated by the underlying model kine-matics.The main purpose of the airway-skin is to facilitate the simulation of bolusinside it (discussed in Section 4.2). In the following sections, we discuss ourframework to generate a deformable airway-skin model to meet the challengesmentioned above and simulate a bouls inside it.4.1.1 Airway geometryThe first step is to create the surface geometry of the airway-skin mesh thatfits the model at the 0th time instant i.e. initiation of swallowing. Autodesk R©Meshmixer R©, a freely available tool for 3D modeling, is used to manuallycreate the airway-skin mesh. The airway-skin is confined by the tonguesurface and hard palate and progress posteriorly towards oropharynx coveringparts of the pharynx. Figure 4.1 show the airway-skin mesh tailored to fitour oropharyngeal biomechanical at the 0th time instant.584.1. Deformable Airway ModelFigure 4.1: Airway-skin mesh coupled with the biomechanical model.4.1.2 Deforming airway-skin with model dynamicsThe airway-skin mesh is attached to the model using ArtiSynth’s attachmentmechanism. In ArtiSynth, different components, like surface meshes, rigidbodies and FEMs, can be attached by making the position qa of an attachedcomponent to be a function of one or more master components qm (detailsare given in [110]). The master component in this case are the dynamic com-ponents of the model i.e. FEMs or rigid bodies, and the attached componentis the airway-skin mesh. The collective state of the attached component isgiven by:qa = f(qm) (4.1)The relation between the velocity of the master component um and ve-locity of the attached component ua is approximated by differentiating withrespect to = Gum (4.2)whereG ≡ ∂f/∂qm (4.3)here qm is the position the master component. ArtiSynth uses Equation4.2 in each step to solve for all dynamic component velocities by applyingEquation 4.1.The airway-skin mesh is attached to underlying dynamic componentsbased on Equation 4.1 by treating each vertex of the mesh as a virtual dy-namic component. These virtual dynamic components are attached to one594.1. Deformable Airway Modelor more dynamic components i.e. mater components. The master compo-nents in this case are 3-DOF FEM nodes and 6-DOF rigid body frames. Theunified skinning method [112] is used to deform the airway-skin mesh withmodel dynamics. The position of each vertex of the airway-skin mesh, qas, isgiven by:qas = qas0 +M∑i=1wifi(qm, qm0, qas0)(4.4)where qas0 is the initial position of the airway-skin point, qm0 is the collectiverest state of qm, wi is the skinning weight associated with the ith mastercomponent, and fi is the corresponding blending function.The bending function fi for a FEM node is the its displacement from itsinitial position qi0 and is given byfi(qm,qm0,qas0)= qi − qi0 (4.5)Figure 4.2: Deformable airway-skin model attached to the dynamic modelcomponents. The deformation of the airway-skin is influenced by the move-ment of surrounding anatomical structures during a swallowing motion.604.2. Bolus Simulation Using SPHWhen considering a rigid body as a master component, the blendingfunction is given for linear and dual-quaternion linear blending. For linearblendingfi(qm,qm0,qas0)= qˆiqas0qˆi−1 − qas0 (4.6)where quaternion product ˆq =(qq0−1)is the relative control frame transformapplied to the initial vertex location qas0. Considering the ith component,the blending function is given asqˆi = qiqi0−1 (4.7)Now for dual-quaternion linear blending fi us expressed asfi(qm,qm0,qas0)= q˜iqas0q˜i−1 − qas0 (4.8)whereq˜i =qiqi0−1‖M∑j=1wjqjqj0−1 ‖(4.9)For more details on duel-quaternion blending please refer to [113]. The re-sulting airway-skin provides a bi-directional coupling that allows forces to betransmitted back and forth between the skin mesh and underlying dynamiccomponents (see Figure 4.2).4.2 Bolus Simulation Using SPHSmoothed Particle Hydrodynamics (SPH) was designed to solve the hy-drodynamic problems represented in the form of partial differential equa-tions(PDE). These PDEs consists of field variables of the fluid (or solid)such as density, velocity, energy and so on. For obtaining analytical solu-tions for the PDEs, the problem domain is discretized at the beginning. Afterthat, approximating of the values of the field functions and their derivativesat each point is assigned. This functional approximation is applied to thePDEs to formulate the ordinary differential equations in a discrete fashionwith respect to time. Now these discretized ODEs are solved with conven-tional finite difference methods. The following list outline the key steps forobtaining analytical solutions for the set of PDEs using SPH. Mathematicalformulation of these steps are discussed in details in the Appendix B.614.2. Bolus Simulation Using SPH1. Meshfree representation: If the problem domain is not representedwith particles, it is described using a set of discrete arbitrarily dis-tributed particles which are not connected by any computational mesh.2. Integral function representation: In SPH methodology integralfunction representation is termed as kernel approximation which is usedfor approximating field function. This mathematically provides the nec-essary stability for SPH method. The integral representation behavesas a weak formulation as it has smoothing effect. Such weak formula-tion make the method robust as long as the numerical integration isperformed correctly.3. Compact support: The kernels are approximated further using par-ticles and termed as particle approximation in SPH methodology. Thisapproximations are achieved by substituting the integral representationof the field functions and its derivatives with summations of field valuesof the neighbouring particles in the support domain (the domain of in-fluence or local domain). This approximation is very important becauseproblems with large deformation requires high resolution simulationswhich means higher number of particles. With support domain, systemmatrices are sparse discretized to help reduce computation effort.4. Adaptive approximation: In each time step, the compact supportfor the particle approximation method is calculated. Therefore, thelocal distribution of the particles contributes to particle approximationand it becomes adaptive. This adaptive approximations at the earlystage of the field variable approximation allow arbitrary particle dis-tribution and ensures that the SPH formulation can handle problemswith large deformations.5. Lagrangian discretization: The particle approximations are appliedto the field functions in the PDEs to formulate a set of discretizedODEs with respect to time. This Lagrangian description allows theSPH method to have all the advantages of Lagrangian descriptionslisted in Table 2.1.6. Dynamic solver: To obtain the time history of the field variablesfor the particles, the discretized ODEs are solved using an explicitintegration technique for achieving fast time stepping. To ensure stable624.2. Bolus Simulation Using SPHtime integration, its important to select correct time step which candepend on the problem description.The aforementioned features make the SPH method an attractive choicefor simulating bolus. Detailed formulation for implementing SPH solver isdiscussed in Appendix B.4.2.1 Artificial viscosity termThe viscosity of a simulated SPH fluid is modeled by a artificial viscosityterm. Hydrodynamic simulation sometime include shock wave that developnonphysical oscillations in the numerical results. Shock wave introduces apropagating disturbance in a fluid when a wave moves faster than the speedof sound in that fluid. Like ordinary waves shock wave carries energy and canpropagate through a medium. The conservation of mass, momentum and en-ergy equation requires simulation of energy transformation from kinetic forminto heat form. This energy transformation can be physically representedas a form of viscous dissipation. This let to the idea of artificial viscosityterm developed called von Neumann-Richtmyer artificial viscosity [114]. Thisvon Neumann-Richtmyer artificial viscosity, Π1 is a quadratic expression ofvelocity divergence, which is given byΠ1 ={a1∆x2ρ(∇·v)2 ∇·v < 00 ∇·v ≥ 0 (4.10)where a1 is an adjustable non-dimensional constant.By adding a liner viscosity term Π2, the oscillation of the numerical resultsare further smoothed that are not damped by the quadratic artificial viscosityterm.Π2 ={a2∆xcρ∇·v ∇·< 00 ∇·≥ 0 (4.11)The above equitation introduces a new terms c, which is called the speed ofsound and a2 which is an adjustable non-dimensional constant. The combi-nation of von Neumann-Richtmyer viscosity Π1 and linear viscosity Π2 arewidely used for removing shock wave related oscillations in hydrodynamicsimulation. Intuitively, the artificial viscosity term spreads the shock wave634.2. Bolus Simulation Using SPHover several cells and regularize the numerical instability caused by disconti-nuity. This term is added to the pressure term to help diffuse sharp variationin the flow and diffuse the energy of the high frequency term.Monaghan and Gingold [107] also developed an artificial viscosity for SPHmethods for simulating shocks and discontinuity. The detailed formulationfor Monaghan artificial viscosity is given in Chapter 4 of [115]. In summarythe Monaghan artificial viscosity Πij is given asΠij =−αΠc¯ijφij + βΠφ2ijρ¯ijvij·xij < 00 vij·xij ≥ 0(4.12)whereφij =hijvij·xij| xij |2 + ϕ2(4.13)c¯ij =12(ci + cj)(4.14)ρ¯ij =12(ρi + ρj)(4.15)hij =12(hi + hj)(4.16)vij = vi − vj and xij = xi − xj (4.17)In the above equation αΠ and βΠ are constants whos values are set around 1.The viscosity associated with αΠ results in bulk viscosity and βΠ is similarto the von Neumann-Richtmyer artificial viscosity. c and v represent speedof sound and velocity respectively. When two particles are approaching eachother, to prevent numerical divergence, ϕ is assumed to be 0.1hij. Theartificial viscosity given by Equation 4.12 is added to pressure terms in theSPH equations.4.2.2 Artificial compressibilityIn SPH formulation the particle motion is driven by pressure gradient. Theparticle pressure is calculated by the local particle density and internal energythrough the equation of state. For the incompressible flows the equation of644.2. Bolus Simulation Using SPHstate leads to extremely small time step which causes numerical instability(discussed in Section 4.2.3). An artificial compressibility term is introducedfor simulating incompressible fluids with numerical stability. The concept ofartificial compressibility assumes that every theoretically incompressible fluidis actually compressible. The artificial compressibility term is introduced toproduce the time derivative of pressure. Monaghan [116] used the followingequation to model free surface flow which is originally the equation of stateof water:p = B((ρρ0)γ − 1) (4.18)where γ is set to be 7 empirically and ρ0 is the reference density. B is aproblem dependent parameter which sets a limit for the maximum change ofdensity. Practically B is taken as the initial pressure [108]. In the Equation4.18, 1 is subtracted to remove the boundary effect for free surface flow.Further this equation also shows that small change in the density results ina large variation of pressure.The artificial compressibility can also be represented by the followingrelationp = c2ρ (4.19)where c is speed of sound. Morris et al. [108] used this equation of statefor modeling incompressible flows using SPH. The SPH simulation resultspresented in this thesis use this relation to enforce incompressibility.The speed of sound is an important parameter for simulating a fluid usingSPH formulation. For example, the actual speed of sound for water understandard pressure and temperature is 1480m/s. If this actual speed of soundis incorporated, a real fluid is approximated as an artificial fluid. Monaghan[116] described the density variation δ asδ =∆ρρ0=|ρ− ρ0|ρ0=V 2bc2= M2 (4.20)where vb is the fluid bulk viscosity and M is the Mach number which is adimensionless quantity representing the ratio of flow velocity past a bound-ary to the speed of sound c in that medium. As the speed of the sound islarge, the corresponding Mach number is very small. As a result, the densityvariation δ becomes negligible and to approximate a real fluid as an artificialincompressible fluid a much smaller speed of sound term in used. However,the speed of sound term should be large enough to make the behaviour of654.2. Bolus Simulation Using SPHthe simulated fluid realistic and small enough that computation time step iscomputationally realistic (see Section 4.2.3). So the speed of sound shouldbe selected in such a way that it balances the time step selection and theincompressible behaviour. Morris et al. [108] considered the balance of pres-sure, viscous force and body force to estimate the optimal value of the soundof speed term described by the following equationc2 = max(v2bδ,υVbδL,FLδ)(4.21)where υ is the kinematic viscosity defined as υ = µ/ρ, F is the body forceand L is the characteristic length scale. The speed of sound term was chosencarefully to maintain the stability of the simulation.4.2.3 Time step selectionSPH formulation reduces the equations of fluid dynamics to a set of ODE ofmotion of the discrete particles. The discrete SPH equations are integratedwith standard methods like Runge-Kutta (RK) method and so on. This stepis computationally expensive which requires several evaluation of the forceterm at each time step.The Courant–Friedrichs–Lewy (CFL) condition is a necessary conditionfor convergence while solving certain partial differential equations numeri-cally by finite differences methods. The explicit time integration schemes aresubject to the CFL condition for stability. The CFL condition states that atime step bigger than some computable quantity should not be taken. Whichin other words mean that the time step must be kept small enough so thatinformation has enough time to propagate through the space discretization.So the numerical simulation should include the physical domain of depen-dence, or the maximum speed of numerical propagation must exceed themaximum speed of physical propagation [95]. The CFL conditions requiresthe time step to be proportional to the smallest particle resolution which inSPH formulation is the smallest smoothing length. The time step ∆t can bedenoted as∆t = min(hic)(4.22)Monaghan [116] suggested the following equation for selecting the time664.2. Bolus Simulation Using SPHstep considering viscous dissipation ∆tcν and external force ∆tf∆tcν = min(hici + 0.6(αΠci + βΠmax(φij))) (4.23)∆tf = min(hifi) 12 (4.24)where f is the magnitude of force per unit mass i.e. acceleration. It can beseen that Equation 4.23 is another representation of Equation 4.22 by addingthe viscous force term. Typically time step is calculated taking the minimumof both of the time step terms using the following equation∆t = min(λ1∆tcν , λ2∆tf)(4.25)here λ1 and λ2 are coefficient which are empirically chosen to be 0.4 and 0.25respectively.Time step is estimated in this Thesis following the formulation of Morriset al. [108] considering viscous diffution∆t = 0.125h2ν(4.26)where ν is the kinematic viscosity defined by ν = µ/ρ.4.2.4 Boundary treatmentBoundary treatment is an important consideration for SPH formulation. Theparticles near the boundary face a problem called particle deficiency whichis due to truncated neighbourhood boundary. When a particles is near theboundary, only particles inside the boundary contributes to the summationof particle interaction even if that particles smoothing kernel expands beyondthe boundary. Such assumptions lead to incorrect solution because on solidsurfaces outside the boundary, all the field variable are not zero. For example,on the solid surface the velocity is zero but density do not necessarily reduceto zero. Monaghan [116] placed line of ghost particles on boundary whichproduces a highly repulsive force to the particles near the boundary. Theseghost or dummy particles prevent the simulated SPH particles to penetratethe solid boundary emulating a solid surface. Some improvement are alsoproposed in simulating the dummy boundary particles.674.2. Bolus Simulation Using SPHTo solve the particle deficiency problem, two different type of dummyparticles are simulated (see Figure 4.3) which are called type I and type IIparticles. Type I particles are similar to the Monaghan type ghost particleswhich are located right on the solid boundary. The type II particles arelocated right outside the boundary. These two types of dummy particlesare specially marked for contribution in the later summation on the SPHparticles. For a simulated SPH particle i which is located within the distanceκhi from the boundary, a dummy particle is placed symmetrically on theoutside of the boundary. So for a SPH particle i near the boundary, all theneighbouring particles within its influence are defined by radius κhi can bedivided into following three categories (see Figure 4.3):Figure 4.3: Illustration of real particles and the two types of virtual particlesused for simulating the solid boundary. Adapted from [115].1. Interior particles: All the SPH particles inside the neighbourhood ofparticle i. These particles are defined as I(i).2. Boundary particles: All the dummy particles simulated right on thesolid boundary that are in the neighbourhood of particle i. These arethe type I dummy particles denoted by B(i). These type of particlestake part in the kernel and particle approximation (see Appendix B684.2. Bolus Simulation Using SPHfor detailed formulation) of the simulated SPH particles. The positionand the physical variables of these dummy particles do not change inthe simulation process. Furthermore, these particles apply repulsiveboundary forces to prevent the SPH particles from penetrating thesolid boundary. When a SPH particle is approaching a type I boundaryparticle, a force is applied pair-wisely along the centerline of these twoparticles. The force PBij is defined asPBij =D[(r0rij)n1 − ( r0rij)n2]xijr2ij(r0rij) ≤ 10, (r0rij) ≥ 1(4.27)where n1 and n2 are selected to be 12 and 4 respectively. D is a problemdependent parameter and chosen at the same scale as the square ofthe largest velocity. rij is the distance between the SPH particle andthe boundary particle in question. The repulsive force is coordinatedby r0 which is defined as the cutoff distance. It effectively chosen tobe equal to the initial particle spacing which means that if a SPHparticle comes closer to the boundary than the initial particle spacing,the dummy particles would apply a repulsive force to establish a solidsurface. Thus the choice of cutoff distance r0 is important because ifits chosen to be to large, some particles in the initial distribution maybe subject to repulsive boundary condition. On the other hand, if thecut off distance r0 is too small, the SPH particle may penetrate thesolid boundary before even feeling the repulsive force.3. Exterior particles: All the dummy particles simulated outside thesolid boundary that are in the neighbourhood of particle i. These arethe type II dummy particles denoted by E(i) which can both be ap-plied to treat solid boundaries and free surfaces. The dummy particleshave the same density and pressure as the corresponding SPH particlewith opposite velocity. The numerical tests [115] have shown that thistreatment of the boundary is very stable and effective. It not only im-proves the accuracy of the SPH approximation in the boundary region,but also prevents the unphysical particle penetration outside the solidboundary.In this Thesis, the SPH boundary particles are simulated as following the694.3. Results and AnalysisUnified-semi-anatytic wall boundary conditions proposed by Ferrand et al[117].As we drive our model kinematically, the SPH particles are only influencedby the movement of the OPAL structures, and not the other way aroundin which the SPH particles can apply forces back and influence the modeldynamics. We are modeling the kinematic trajectories of the 3D surfaces asthey move, rather than modeling the influence the fluid has on the dynamicstructures.4.3 Results and AnalysisThe kinematically driven biomechanical model developed in Chapter 3 wasbuild using ArtiSynth ( [110]. The SPH simulation is alsoperformed using ArtiSynth with the formulation described in [118]. The SPHparticles carry estimated physical quantities, e.g. velocity, density, pressure,by approximating values and derivatives of discrete samples of field quanti-ties. These particles are analogous to particles occupying a fraction of thefluid domain and produce accurate results if the particles are dense i.e. highresolution.4.3.1 Simulation of a stable bolusThe Navier-Stokes equations for a incompressible Lagrangian system can bewritten as:DvDt= −1ρ∇p+ 1ρ(∇ · µ∇)v + B, (4.28)DρDt+ ρ∇ · v = 0, (4.29)where v is the velocity, ρ is the density, p is the pressure, µ is the New-tonian viscosity of the fluid and B is the body force per unit mass. We solveEquations 4.28 and 4.29 explicitly with an artificial speed of sound term usingSPH approximations. Here the body force B is generated by the dynamicsof the swallowing model i.e. movement of the tongue during swallowing. Theforce is propagated through the deforming boundaries of airway-skin mesh704.3. Results and AnalysisFigure 4.4: Comparing simulation results: Top row VF frames of a normalswallow, middle row 2D swallowing animation frames (bolus shown in white),and bottom row simulated SPH fluid bolus (in blue).(discussed in Section 4.1) and is applied to the SPH particles simulated in-side. Velocity, v, of each particle is solved for each time step and the positionof each particle is advanced accordingly.The airway-skin mesh (derived in Section 4.1) acts as the boundary forthe SPH formulation. SPH bolus particles are simulated inside the deformingairway-skin mesh. The boundary particles are simulated at the surface of theairway-skin mesh and the exterior particles are simulated outside the solidboundary of the surface of airway-skin mesh. These special particles enforcesboundary conditions for the SPH simulation. These particles apply repulsiveforces to the SPH particles that come closer to the boundary more than theinitial inter-particle spacing. As a result the SPH particles move to representthe bolus movement with the deforming airway-skin mesh. The simulationis advanced using small time steps to maintain simulation stability.Typically during a swallowing examination, patients are presented with714.3. Results and Analysisa 5-mL teaspoon of bolus. The simulated bolus was made to emulated thephysical properties of a real bolus used during an MBS examination. TheSPH particles are simulated with an initial inter-particle spacing of 2 mmwith 621 fluid particles simulating about 5 mL of fluid inside the deformingairway-skin mesh. The viscosity µ, of the fluid was set to be 1000 cen-tipoise(cP), similar to a thick fluid like honey with the density ρ initializedat 1000 kg/m3. We applied a gravitation force from a superior to inferiordirection, equivalent to someone swallowing in an upright position.Figure 4.4 compares the instances of our simulation with the correspond-ing frames of VF and animation. The results show that our model is capableof simulating a bolus in a manner consistent with the input data and matchthe swallowing kinematics identified in the animations. The airway-skin wasable to contain the bolus and allowed the model dynamics to apply forces onthe SPH particles. Even though, we did not explicitly define a soft plate inthis model, the airway skin provided the necessary boundary to restrict thefluid particles from entering the nasopharynx.4.3.2 Simulating bolus with standardizedconsistenciesDiet adjustments are sometimes recommended for dysphagia patients to helpmaintain their nutritional needs. For instance, thinner liquids like water maybe difficult to swallow safely if the patient has a delayed pharyngeal swal-low or oral motor impairment [18]. Adjusting the diet by thickening the thinliquid may alleviate the swallowing impairment and minimize chances of aspi-ration. Clinical observations [119; 120] support the hypothesis that increasein bolus viscosity increase patient effort with positive effects on tongue baseand pharyngeal wall movement in some dysphagic patients. It is presumedthat increase in these clearance mechanisms are related to sensorimotor adap-tations. We tested our model’s capability of simulating fluid with differentconsistencies (i.e. viscosity) by simulating 4 boluses with different dynamicviscosity each representing standardized bolus consistency in context withdysphagia. The standardized bolus consistencies instituted by the NationalDysphagia Diet [111] task force are as follows:• Thin, 1050 centiPoise (cP)• Nectar-like, 51–350 cP724.4. Summary• Honey-like, 351–1,750 cP• Spoon-thick, >1,750 cPWe simulated each type of bolus with the same set of solid boundarymotions provided by the airway-skin mesh. Qualitative observations indi-cate that the thinnest bolus (i.e. water-like) escapes the oral cavity with thegreatest and the thickest one (i.e. spoon-thick) with the least velocity, andthe intermediate ones follow the sequence. This result is intuitive since weexpect our model to simulate significantly different bolus positions and trackthe mass of the bolus, when the viscosity of the fluid is changed for the sameset of kinematics.4.4 SummaryWe demonstrated that our model is capable of simulating SPH bolus stablyand is capable of simulating boluses with different viscosity. This sets up theground work for investigating the influence of bolus on structural movementwith a modeling based approach. Further refinement of our model will in-clude simulation of aspiration risk related to abnormal changes in dynamicswallowing movement(s) for different volumes and viscosity. Our future di-rections will include development of models that predict the effect of targetedtreatment(s) on the dynamic movement of swallowing structures and airwayprotection. Model development of normal swallowing and dysphagic patientspecific impairment models are first steps toward this long range goal.73Chapter 5Extending StandardizedDysphagia Training MaterialsIn the previous two Chapters, we developed a full 3D biomechanical modelof the oropharyngeal complex (see Chapter 3) and simulated a fluid bolusdriven by the kinematics of the model. This chapter discusses about the userstudy conducted in this Thesis. The study was deigned to assess, evaluate,quantify and gather feedback from the expert clinicians about the additionalvalue our model can add to the clinician training for standardized dysphagiadiagnosis. The first section of this chapter identifies the clinical collaboratorswho helped to carry out the user study. Following that, the two phase userstudy is discussed in details in the later sections. Finally, at the end of thischapter, the findings of the study, recommendation from the expert cliniciansand concluding remarks are made.5.1 Clinical CollaboratorsDr. Bonnie Martin-Harris from Medical University of South Carolina (MUSC),SC, USA provided the videofluoroscopy data used in the MBSImPTM c© on-line training platform as a part of our ongoing collaboration. We used thisdata to formulate our study design and conduct a pilot stusy. This datasharing met the minimal risk human ethics application criteria and thus anexpedited review by the UBC Clinical Research Ethics Board(CREB) wasconducted (UBC CREB number H15-02749).5.2 Study DesignThe study was divided into two major phases. During the first phase we con-tacted the SLPs from the local affiliated hospitals and asked to meet themfor an interview. During this interviews, the SLPs were asked if extension745.3. Phase I- Interviewing the Stakeholdersof the training materials for MBSImPTM c© i.e. additional 3D visualizationcan contribute to standardized training of dysphagia. Based on the feedbackfrom the first phase we conduct a formal user study to access if enhancedvisualization provided by 3D biomechanical models add value in cliniciantraining for standardized dysphagia diagnosis. This user study met the min-imal risk human ethics application criteria and thus expedited review bythe UBC Behavioural Research Ethics Board (BREB) was conducted (UBCBREB number H15-02665).5.3 Phase I- Interviewing the StakeholdersThe aim of this phase of the user study is to do a feasibility analysis ofour developed biomechanical modeling approach for complementing the VFand animation video used in standardized dysphagia diagnosis training. Theprimary stakeholders are the Speech Language Pathologists. A series of inter-views were conducted to collect feedback from SLPs to justify the feasibilityanalysis. All of the clinicians interviewed during this phase were registeredMBSImPTM c© registered professionals which means after finishing the train-ing, they passed the reliability testing with 80% or higher. The reason behindconducting the interviews were three fold.1. To evaluate the effectiveness of current state-of-the-art dysphagia train-ing.2. To analyze the role of visualization in swallowing assessment examina-tion.3. To identify areas of swallowing continuum where biomechanical mod-eling can provide additional visualization and flexibility for assistingdysphagia training.SLPs affiliated with St. Paul’s Hospital, Vancouver, BC, Canada andRichmond Hospital, Richmond, BC, Canada was interviewed during thePhase-I of the study where we conducted series of interviews were conducted.When we interviewed the SLPs, we asked the following question:• When did you take MBSImPTM c© course?• Why did you take this course?755.3. Phase I- Interviewing the Stakeholders• How does MBSImPTM c© help you professionally?• What do you think was the most important part of the training?• Do use the MBSImPTM c© 17 component grading system in your currentpractice?• How has the training improved/changed/not-changed your practice?• What swallowing impairments are easy to score?• What swallowing impairments are hard to score?• In your opinion, what type of visualization for training would help tobetter comprehend swallowing in order to evaluate dysphagia?• What features did you wish existed in existing training system?• What improvements would you love to have?The first few questions were set to get an idea about SLP training. Thesequestions helped us better understand the importance of swallowing impair-ment standardization and how state-of-the-art training is helping the SLPsfollow the standardization in real practice. This discussion set up the stagefor the discussion about the research question that “can additional visualiza-tion / tools add value to the training”. While discussing about the followingquestions, we wanted their expert opinion to justify the need for additionalvisualization (3D perspective view) and flexibility (simulating different bolusconsistencies) in dysphagia training. The feedback from the interviews aresummarized under the following headings:Standardized MBSS as a tool for swallowing assessmentAll participating clinicians acknowledged that there are several techniqueslike MBS, FEES or bedside swallowing assessment for evaluating a dysphagicswallow. However, they also acknowledged that an MBS examination is pre-ferred because it is standardized with the MBSImPTM c© protocol. After anMBS study is performed and scored using the MBSImPTM c© protocol, theSLPs recommend diet modification, tongue exercises and/or other proceduresto alleviate the swallowing abnormality. The participating SLPs mentioned765.3. Phase I- Interviewing the StakeholdersFigure 5.1: Example of videofluoroscopic and animated images used dur-ing MBSImPTM c© web-based training program. Image reproduced fromMBSImPTM c© online training [76] with permission.that such standardization helps them teach new SLPs about accessing differ-ent swallowing impairment. They also mentioned that MBSImPTM c© stan-dardization makes it easier to communicate the patient swallow scores andenabled them to keep track of patient improvement quantitatively.The clinicians were also asked about their experience with the MBSImPTM c©online training platform. Figure 5.1 shows the existing user interface of theonline training. The participating clinicians registered for a seminar followedby the online course itself. The seminar is designed to give the trainee SLPsorientation about the MBSImPTM c© protocol. After that the SLPs starteddoing the formal online training followed by reliability testing. The partici-pating clinicians appreciated the structured nature of the course and foundthat the swallowing animations used in training were able to illustrate thesalient features of a swallow sufficiently to discern between a normal and im-paired swallow. However, the participating clinicians also pointed out thatsometimes the movement of the bolus and corresponding anatomical struc-tures are not realistic. It is important to note that the bolus and structuralmovements in the animation are unrealistic as they are not bounded by thelaws of physics even if they are fine tuned to match the VF of a patient swal-775.4. Phase II - User Studylow. They suggested that if the additional visualization provided with theVF during training were more realistic, it might contribute in the cliniciantraining.Most difficult physiologic swallowing component to scoreThe participating clinicians were asked to pick three components that theyfind easiest and hardest to score using current training protocol. The com-ponents the clinicians thought were easiest to score were: Component 1 - lipclosure (Lip C), Component 2 - hold position/tongue control (HP), Com-ponent 9 - anterior hyoid motion (HM). The clinicians listed Component 4- Bolus transport/lingual motion(BT), Component 5 Oral Residue, Com-ponent 13 - Pharyngeal Contraction (PC) as hardest to score. This line ofquestioning allowed a discussion to figure out why some components are eas-ier to score and why some are not, and if the visualization used during thetraining contributed to it. Furthermore, this discussion set up the stage foraddressing the need of additional visualization in the current MBSImPTM c©training protocol.5.4 Phase II - User StudyThe clinicians with MBSImPTM c© certification are required to have ≥ 80%accuracy in training to be considered reliable. During field observations byour collaborators who instituted this protocol, it was found that some swal-lowing components are harder to score accurately than others. To identify thecomponents with greatest level of difficulty we analyzed the MBSImPTM c©reliability test metrics in September 2015 [121]. This included 7441 test ses-sions 18328 individual component scores scored by 3461 trainees since thecreation of the database in 2011. It was found that the physiologic compo-nents of the oral domain are most difficult to score. Furthermore, Component4- Bolus Transport/Lingual Motion showed 55% accuracy in all test sessionmaking it the most difficult component to score.The explanation behind this may be the oral phase of swallowing is largelycontrolled by voluntary movements of oropharyngeal muscle and structures.For example, tongue motion is largely controlled by voluntary motion whichhas a high variability among subject population. Due to such high vari-ability the oral components are sometimes harder to score in a standardized785.4. Phase II - User Studymanner. On the other hand, although pharyngeal phase of swallow is morecomplicated and require intricate coordination, pharyngeal phase of swallowis largely controlled by involuntary muscle and structures. As a result, vari-ability in patient population is lesser compared to that of oral components.The state-of-the-art standardized dysphagia training module, MBSImPTM c©,includes videofluoroscopy (VF) and corresponding 2D animations of swallow-ing. However, during Phase I of the user study and the reliability data fromMBSImPTM c© online training platform suggests that clinicians have mostdifficulty scoring oral components, particularly Component 4 – Bolus Trans-port/Lingual Motion, even though some pharyngeal components are oftenregarded as most complicated to comprehend. The training materials avail-able as a aid for scoring the Component 4 and other oral components arecaptured from lateral mid-sagittal cut plane. So we concluded that trainingmaterial itself might be a factor contributing to the complexity of scoringoral components.5.4.1 MethodologyThe goal of this study is to investigate if additional 3D visualizing duringclinician training learning standardized dysphagia diagnosis, adds value inthe training, and help the trainees to appreciate the swallowing anatomyand physiology in three dimension to better detail the physiologic charac-teristics contributing to dysphagia. So it is hypothesised that additional 3Dvisualization showing the movements of the oropharyngeal structures duringa swallowing motion are likely to help clinicians understanding of swallowingdynamics and allow them to score more confidently following standardizeddysphagia diagnosis protocol. We selected Component 4 as it has the low-est scoring accuracy to test in our user study. We wanted to know if ourassumption that extending the training material by adding 3D perspectivevisualization can have positive impact and help the clinicians distinguishbetween different impairments of component 4.Component 4 - Bolus Transport/Lingual Motion is evaluated at the onsetof productive tongue movement to propel the bolus through the oral cavity.Component score definition for Bolus Transport/Lingual Motion are listedbelow:(4–0) Brisk tongue motion – The movement of the tongue is briskand timely without any hesitation that results in good bolus transport.(4–1) Delayed initiation of tongue motion – To transport the bolus795.4. Phase II - User StudyFigure 5.2: For a sore of 3 of Component 4, the tongue and the bolus movesback and fourth within the oral cavity before moving the bolus into thepharynx. On the other hand, if the bolus is slowly manipulated but progressesthrough the oral cavity in a single direction toward the pharynx, it is a scoreof 2 for Component 4.through oral cavity, the patient requires multiple cues to initiate movementof the tongue. However, once initiated, the bolus progresses normally.(4–2) Slowed tongue motion – The tongue motion is slow and weak.Despite the slow tongue movement, the tongue provides a productive poste-rior ward motion to propel the bolus through the oral cavity.(4–3) Repetitive/disorganized motion – The tongue move back andforth moving the bolus anteriorly and posteriorly before transporting thebolus through oral cavity.(4–4) Minimal to no tongue motion – There is minimal or no tonguemotion despite cuing.With respect to Component 4, a score of 0, 1 and 4 are usually apparent.This is because brisk, delayed or minimal tongue motion is usually easier805.4. Phase II - User Studyto identify. However, while differentiating between a score of 2 and 3, it issome what more difficult to discern between slowed and repetitive tonguemotion (see Figure 5.2). This user study is aimed to evaluate if the addi-tional visualization and flexibility provided by our 3D model can add valueto the SLPs training learning standardized dysphagia diagnosis protocol. Toevaluate the added value, we built an interface similar to the training zonefor MBSImPTM c© online training platform and extended the existing train-ing material available for scoring Component 4 - Bolus Transport/LingualMotion. With the interface we performed a user study involving SLPs wherethey scored with the additional visualization provided for component 4.We used our model to generate 3 different views from prospective 3Dviewpoints for all five scores of Component – 4. These viewpoints wereselected to enhance the salient feature of the individual components scoresand were carefully selected and reviewed by our expert clinical collaborators.5.4.2 Participant selection criteriaStudy population for this study is was narrow due to the selection crite-ria and feasibility to conduct the study. We wanted to recruit SLPs withMBSImPTM c© experience for participation. So we contacted the SLP de-partment of the major health-care facilities located Vancouver, BC, Canada.In the scope of this Thesis, only recruiting SLPs in the Vancouver was feasiblefor logistics issues. We advertised about the user study across health-carefacilities in Vancouver, BC, Canada. We recruited participants from St.Paul’s Hospital, Richmond Hospital, Lions Gate Hospital, GF Strong Reha-bilitation Centre and Royal Columbian Hospital, and were able to recruit 8participants. Although the there were limited number of participants in theuser study, the expert opinion of the SLPs helped used evaluate if there wasany additional value added to the training by the 3D views.5.4.3 User interfaceWe built an interface that is similar to the existing “swallow by swallow” sec-tion in training zone of MBSImPTM c© protocol (see Figure 5.1). We included10 MBS studies in the interface and asked the participants to score the first4 components of the MBSImPTM c© for each of the 10 cases. The first 4 com-ponents are Component 1 – lip closure, Component 2 – tongue control duringbolus hold, Component 3 – bolus preparation/mastication and Component815.4. Phase II - User StudyFigure 5.3: Additional 3D views to component the existing training materialsfor Component 4 was added to the new user interface built for the user study.4 – bolus transport/lingual motion. While scoring Components 1 through3, the participant had the option to review the 2D video correct componentscore animation in case of an incorrect score. This functionality is availablein the existing training platform (see Figure 5.3). So the users had accessto similar training materials when they were scoring the first 3 components.For component 4 Bolus transport/lingual motion, we included 3 different 3Dviews called 3D oblique tongue view, 3D A-P tongue view and 3D S-I tongueview in case of an incorrect score. Figure 5.4 shows the same instances of VFwith the 3 additional 3D views added as an extension of training materialavailable for scoring normal swallow motion for Component 4. Figure 5.5makes the same comparisons for an impairment 4-3 (Repetitive/disorganizedmotion) without the bolus.The study interface was designed in a way that the additional visualiza-tions for Component 4 is only available when an incorrect score is selected.So if a participant scoreed all the components correctly he/she would nothave been exposed to the additional 3D views. To make sure that clini-cians are able to give feedback, an interview session was conducted afterthe participants completed their task and before they started completing thequestionnaire. During the interview, the participant were shown different 3D825.4. Phase II - User StudyFigure 5.4: Top row: An MBS case for a patient with a score of 0- Brisktongue motion tongue motion for Component 4 - Bolus Transport/LingualMotion. The bottom three rows show 3D oblique tongue view, 3D A-P tongueview and 3D S-I tongue view respectively which was added to the traininginterface to extend the existing training material for scoring Component 4.views and asked which view would help them identify the salient features todiscern between different impairment of component 4. By doing so, we madesure that all the participants are exposed to the 3D views and be able to giveus feedback on the views.835.4. Phase II - User StudyFigure 5.5: Top row: An MBS case for a patient with a score of 3- Repeti-tive/disorganized tongue motion for Component 4 - Bolus Transport/LingualMotion. The bottom three rows show 3D oblique tongue view, 3D A-P tongueview and 3D S-I tongue view respectively which was added to the traininginterface to extend the existing training material for scoring Component 4.QuestionnaireAfter finishing the study, the participants were asked to complete a question-naire. The questionnaire is designed to identify if the additional visualizationafforded by our model provided useful information to clinicians learning thestandardized dysphagia diagnosis protocol. The questions were set up to ac-cess, record and quantify the response from the participants as they are thestakeholders as well as experts in the field.845.4. Phase II - User StudyThe first part of the questionnaire consisted of questions about gender,age, years of experience as a SLP and affiliated hospital/health-care facility.We collected these information to help us analyze the data. Category typequestions were set for indicating their gender. Ordinal questions were set forage and experience where the responses were recorded in categories. The agecategories were set as 19–25, 26–60, 31–40, 41–50, 51–60 and 61 or aboveyears. For experience, the categories were 0–3, 3–5 and more than 5 years ofexperience.The next part of the questionnaire consisted of 6 Likert type questions[122] using a five point scale evaluation. A Likert scale is a sum of responsesto several Likert items. A Likert item is a statement where the participant isasked to evaluate the statement with a quantitative value. The evaluation isusually subjective with level of agreement and disagreement with the state-ment. We used a typical five point Likert item with the following values toindicate the level of agreement and disagreement:• SD – Strongly Disagree (1)• D – Disagree (2)• N – Neutral (3)• A – Agree (4)• SA – Strongly Agree (5)We formulated the questions in a way to evaluate and record the cliniciansresponse after they are exposed to 3D views provided by our model. Theseparticipating clinicians previously have undergone the standard training, sothe questions were also set to see if they think these additional features couldadd value to the training. The questions set in the questionnaire are listedbelow:Q1 I revisit the example videos in the MBSImPTM c© online training plat-form during my normal practice.Q2 I find the example videos in the MBSImPTM c© online training platformuseful.Q3 I was able to easily switch between the different 3D views which wereavailable for scoring Component 4 - Bolus Transport/Lingual Motion.855.5. Results and analysisQ4 The 3D views available for scoring Component 4 - Bolus Transport /Lingual Motion, provided additional information that was not providedby the videos.Q5 The 3D views were useful in differentiating between different impair-ments of Component 4 - Bolus Transport / Lingual Motion.Q6 In my opinion, I think 3D views would be a helpful learning material fortrainees.5.5 Results and analysisFigure 5.6: Variation of age and experience in the study population.The aim of the user study was to gather feedback on adding features in thetraining zone that is provided by our 3D model. There were 8 participantsin the study and all of them were women coincidentally. The participantswere affiliated with health-care facilities/ hospital located in Vancouver, BC,Canada as listed in Section 5.4.2. Majority of the participating clinicianswere aged between 31–40 (see Figure 5.6) and had more than 3 years ofexperience (see Figure 5.6). The following sections analyzes the5.5.1 Questionnaire analysisThe next part of the questionnaire was the Likert scale questions which areseparated into 3 groups for better analysis and visualization. Group I con-sisted of Q1 and Q2. In these questions we wanted to know if the participant865.5. Results and analysisFigure 5.7: Median and Mode of the questionnaire response score for thestudy population (n=8).revisit the training materials available in the existing system and find themuseful. The average agreement for the first two question were respectively2.375 and 4.33 out of 5. Group I questions identified the participants whoare likely to be influenced by the additional training materials the most. Theaverage agreement score in these questions implies that the participants areup to date about the training materials and can comment on if extension oftraining material is needed for enhancing the training.Q3 is listed in Group II as it asks if the participant could switch betweendifferent views easily. This question is related to the user interface. Addingadditional features to a interface sometimes adds complexity and have neg-ative influence on the user experience. The response to this question helpedus to evaluate if the added features introduces a level of complexity thathas minimal impact on user experience. The average agreement score of thisquestion is 5 out of 5 which means all participant strongly agree that theycould switch between different views without having any negative impact onthe user experience.Group III consisted of Q4, Q5, Q6. These questions evaluated the partic-ipant’s response about the usefulness of the additional 3D views. Q4 asked875.5. Results and analysisFigure 5.8: Range of the questionnaire response score with mean for thestudy population (n=8).if the participant thought that the 3D views provided additional informationto evaluate Component 4 and the next question asked if these additionalinformation were useful. The average agreement score these questions wererespectively 4.67 and 4.33 out of 5. The last question, Q6 asked the par-ticipants if they thought the 3D views would be an useful addition in theexisting training system. The average agreement score for this questions was4.375 out of 5 which means that the clinicians think that the participatingclinicians believed that visualizing a swallowing motion in 3D cloud help theclinician training and contribute to extend the training materials availabletoday.Figure 5.7 shows the median and mode of the questionnaire responseof the study population. Figure 5.8 shows the minimun and maximum ofthe questionnaire response of the study population. It can be seen that theresponses for the Group III questions hold a high level of agreement score.This data from our pilot study support our initial assumption that adding3D model in the standardized for that the additional visualization mightbe an useful extension for scoring and learning the standardized dysphagiadiagnosis protocols.885.5. Results and analysis5.5.2 Testimonial dataAfter the participants completed the questionnaire, they were asked if theywanted to give any additional feedback about the 3D views and their useful-ness. Their testimonial data from the user study suggest the following:Additional 3D views provide useful informationAll the clinicians who participated in the study agreed the 3D view providedadditional information about the swallowing swallowing dynamics. Theyalso agreed that the additional information afforded by our 3D model can beuseful for differentiating between the salient feature of Component 4; theyagreed with an average agreement score of 4(see Figure 5.7). They reportedthat viewing the tongue movement in 3D form different perspective viewspoints helped them to identify its direction of motion and provided themmore detail about the physiology.3D oblique tongue view is most useful for scoring Component 4For scoring Component 4, the participating SLPs found the 3D obliquetongue view most useful. This view is similar to the mid-sagittal cut planeview only from a perspective viewpoint and with a full 3D model of tongue.In practice and training, the SLPs need to infer the tongue motion from 2Dvideofluoroscopic images. They mentioned that the 3D oblique tongue viewallows them to view full tongue motion in a manner that is similar to lookinginside the mouth of the patient during a swallowing motion. With this view,they agreed that they were able to more clearly discern between differenttongue movement pattern descriptive of different Component 4 impairments.After the 3D oblique tongue view, the participating SLPs thought that 3DA-P tongue view provided useful information. Even with a anterior-posteriorVF scan, movement of the tongue tip is not readily visible. The existingtraining materials also do not illustrate the tongue tip motion clearly as itshows the swallowing from the midsagittal cut plane. The SLPs thoughtthat the 3D S-I tongue view (see Figure 5.4 and 5.5) could provide moreinformation identifying Component 5 - Oral Residue and other pharyngealcomponents which are best viewed from superior-anterior viewpoint.895.5. Results and analysis3D views might be useful extension in MBSImPTM c© trainingAll the participating clinicians recommended with an average agreementscore of 4.375 (see Figure 5.8) that 3D views can be a useful extension ofthe training materials for MBSImPTM c©. During the interview, we asked theexperts to recommend which physiologic components of swallowing wouldbenefit most from adding 3D extension to the existing training materialsas additional scoring aid. After conducting a series of interviews with theparticipating SLPs, we discussed our findings and their recommendationswith our clinical collaborator Dr. Bonnie Martin-Harris and her team at theMedical University of South Carolina (MUSC), Charleston, SC, USA. Dr.Martin-Harris and her team at MUSC instituted the MBSImPTM c© proto-col and developed the existing training materials to help the clinicians scorereliably following the standardized protocol. Dr. Martin-Harris agreed thatthe 3D views provided by our model might help extend the existing trainingmaterials and help the trainees clarify concepts about different componentscores.5.5.3 RecommendationThe participating SLPs from Vancouver, BC, Canada and Dr. Martin-Harris’s team at MUSC recommended that the following components mightalso benefit from 3D extension of the existing training materials:Component 15 – Tongue Base RetractionComponent 15 – Tongue base retraction is scored at the maximal retractionof the tongue base. Observations of tongue base retraction are made basedon the presence and degree of bolus or air between the base of tongue andposterior pharyngeal wall (see Figure 5.9). It is view from a mid-sagittalplane provided by an MBS study. The retraction of the tongue base resultsin a “merging” of the base of the tongue with the posterior pharyngeal wall.So score of 0 to 4 are made judging the interaction between the base ofthe tongue and posterior pharyngeal wall. It is somewhat challenging tocomprehend the interaction between these structures from 2D videos. Thus,the experts recommended 3D views might make significant difference if addedto the existing training materials.905.5. Results and analysisFigure 5.9: Observations of tongue base retraction are made based on thepresence and degree of bolus or air between the base of tongue (outlined inred) and posterior pharyngeal wall (outlined in blue). VF image reproducedfrom MBSImPTM c© training module, with permission.Component 8 – Laryngeal ElevationComponent 8 – Laryngeal Elevation is scored at the time the epiglottisreaches a horizontal position and it is scored just after initiation of hyoidmotion which signals the onset of pharyngeal swallow. Hyoid excursion andthe rise of the larynx sometimes cannot be visualized clearly in a VF andthe animated illustrations in the existing training module. Furthermore,relative movement of thyroid cartilage, epiglottic petiole and arytenoid car-tilages plays an important role when deciding between different scores of thiscomponent. So the experts suggested if the existing training materials areextended with 3D views to evaluate “minimal” to “partial” laryngeal eleva-tion, the trainees will be able to comprehend the salient features pertinentto this component scores.915.6. DiscussionComponent 13 – Pharyngeal ContractionComponent 13 – Pharyngeal Contraction is scored from a A-P view andthis components is scored as a combination of pharyngeal shortening andstripping wave. For a score of 0 the pharyngeal shortening is symmetric andthe contraction is complete. If there is unilateral bulging of one pharyngealwall the score is 2 and for bilateral bulging of both pharyngeal walls thescore is 3. For a case where the contraction is incomplete the score is 1. 3Dviews provided by our model can help determining the completeness of thepharyngeal contraction or the symmetry of the shortening of the pharynx,which cannot be easily seen from VF examples or the animated illustrationcaptured from the AP view. The clinicians agreed that adding 3D views tothe training of this component might have positive impact on the traineesunderstanding of this component.5.6 DiscussionThe pilot user study was conducted as a qualitative feasibility analysis toevaluate if the 3D perspective visualization provided by our model cab bean useful extension to the standardized training for dysphagia diagnosis.One limitation of this study was having small study population. We werelimited by our resources so we only recruited SLPs in the greater Vancouverarea with MBSImPTM c© certification. This limited our potential participantpool. However, to determine if scoring accuracy increases with 3D extensionprovided by our model, we would have to recruit more participants andcompare their scoring efficiency of one group trained with and one grouptrained without the 3D extended visualization.5.7 SummaryTo summarize, we conducted a user study involving expert clinicians to in-vestigate if the additional flexibility and visualization provided by our modelcan enhance the existing training materials for MBSImPTM c© protocol. Thisstudy involved building a training interface for adding the 3D extension to theexisting training material, recruiting clinicians for participating in the study,analyzing data from the questioner and testimony. The result from the pi-lot data indicate that participating clinicians believe that the additional 3D925.7. Summaryviews are useful for identifying the salient features for differentiating betweendifferent swallowing impairments, such as direction, strength and timing ofthe tongue motion, and could be a useful addition to the current standardizedMBSImPTM c© training system.93Chapter 6ConclusionsTo conclude, this chapter reviews and summarizes the contributions of thisdissertation in terms of impact and potential application. It also acknowl-edges the remaining challenges and point towards the long term goals drawingon this research.6.1 Impact and Potential ApplicationThe contributions of this dissertation include a modeling frame work for gen-erating a 3D biomechanical swallowing model from animated illustrations, astable bolus simulation driven by the model kinematics and an extensionof training materials used for standardized dysphagia training. For devel-oping our model, we undertook participatory designing approach involvingall stakeholders. The stakeholders in our case included the clinical educa-tors who instituted standardized dysphagia diagnosis protocol and the SpeechLanguage Pathologists who are learning/training/using the standardized pro-tocol. We involved both type of stakeholders in our design process incorpo-rated their feedback to produce what was needed to enhance the existingtraining materials. The contributions of this dissertation are summarizedbelow:Modeling of oropharyngeal swallowing motionOropharyngeal geometries are complex and consist of bony and soft anatom-ical structures. Generating a realistic swallowing model with accurate timingof swallowing event is a challenging task even with state-of-the-art imagingtechniques; various research groups from different domains are trying to an-swer this challenge. We took a bottom up approach to build our model fromthe animation models used in clinician training. These animated illustrationsare created by expert clinicians, using their knowledge about oropharyngeal946.1. Impact and Potential Applicationanatomy and observing real swallowing events using different imaging modal-ities. These animated illustrations are validated for clinicians training, thusthese animations incorporate very important kinematic information of a swal-lowing motion.We built a 3D biomechanical model of the oropharyngeal complex fromrealistic geometries derived from animated illustration used in clinician train-ing. Sufficiently complex and realistic geometry is important for a biome-chanical model to perform relevant and meaningful simulation. We generatedrigid body models for bony structures and finite elements models for softermore deformable structures of the oropharyngeal complex. We coupled therigid body and finite element models, and incorporated accurate and vali-dated timing of swallowing events from the training materials available forstandardized dysphagia diagnosis protocol to drive our model. With thisapproach we are able to bridge the gap between the 2D swallowing motionviewed in an MBS study and complement it with a more detailed 3D oropha-ryngeal swallowing model. For example, normally, in–vivo it is difficult tosee the full dimension of tongue base retraction, supraglottic airway closureand particularly contraction of the pharynx, and cannot assume symme-try of movement from a lateral view. Our model allows for such detailedvisualization and deeper comprehension of swallowing dynamics.Simulating a bolus driven by model kinematicsSimulating a fluid bolus is a challenging computational fluid dynamics prob-lem as it involves many level of complexity. The dynamics of the soft andbony oropharyngeal structures provide boundaries for the bolus which drivesthe bolus through the oral cavity. It is important to track the bolus to de-termine if there is any oral or pharyngeal residue, or to detect the presenceof aspiration.We used unified geometric skinning approach to incorporate the defor-mation of the upper airway during a swallow motion. This deforming airwaymesh acted as the deforming boundary for the bolus simulations. We sim-ulated a stable fluid bolus in the airway-skin driven by model dynamics toemulate oropharyngeal swallowing motion in our model. We demonstratedthat our model can simulate a bolus in a manner consistent with the VF andanimated illustrations used in standardized clinician training for dysphagiadiagnosis. Being able to simulate a bolus driven by model dynamics couldallow the clinicians to change the timings of the movements of oropharyngeal956.2. Future Directionsstructures to emulate a swallowing impairment, and visualize the resultingchanges such as the bolus flow, presence of residue or aspiration. We alsodemonstrated that our model can simulate boluses with different viscosityby simulating standardized bolus consistencies used in clinical practice ofdysphagia diagnosis. This marks our contribution by setting up the groundwork for investigating the influence of bolus on structural movement with abiomechanical model based approach.3D extension of training material for standardizeddysphagia diagnosisOur biomechanical swallowing model provides 3D perspective visualizationthat can be used as an extension of the standardized training materials cur-rently used in MBSImPTM c© online training platform.We conducted a pilot user study involving expert clinicians who are theintended users and stakeholders, to assess, evaluate, quantify and gatherfeedback about the additional value that our model might add to standard-ized dysphagia diagnosis. The pilot data from the user study indicate thefollowing:Firstly Visualization of swallowing motion in 3D from perspective view-points afforded by our model, can provide complementary informationfor scoring the most difficult physiologic component for standardizeddysphagia diagnosis.Secondly Perspective 3D visualisations afforded by our model might al-lows the clinician to understand, evaluate and quantify salient featuresspecific to a particular swallowing impairment in more detail duringstandardized training.Thirdly Extension of training materials in the existing system with 3Dviews provided by our model might improve the standardized train-ing outcome.6.2 Future DirectionsThis work sets up the ground for some promising research directions. Shortterm research directions for each contribution are discusses at the end of966.2. Future Directionscorresponding chapters. Some potential long term research directions arelisted as the following:Transition from model space to patient spaceIn the scope of this Thesis, we build our model enforcing symmetry, becausethe aim of this work was to investigate the feasibility of using SPH to simu-late a fluid bolus in a physics-based 3D model that is driven by kinematicsderived from clinical data. Our model can be extended to allow for patientspecific geometries and timings of swallowing events. With these patient spe-cific inputs, our model could generate patient specific swallowing simulationswhich may lead to improved identification of physiologic treatment targetsby testing the effectiveness of intervention strategies on the model ratherthan trial and error in-vivo testing.Treatment alternativeA potential application for our oropharyngeal biomechanical model is as atool for treatment planning and through quantitative analysis of trade-offsbetween treatment alternatives. For example, in case of head and neck can-cer, sometime its required to do resection or reconstructive surgery. Withaccurate, validated, and patient-specific biomechanical models, the functionaloutcome of the surgery can be predicted. Alternative treatment pathwayscan be evaluated quantitatively to predict surgical effect on swallowing abilityto choose the best surgical outcome in term of quality of life.Muscle driven oropharyngeal modelArtiSynth supports a muscle driven FE model where the muscle fibres em-bedded in the model can be activated to follow a desired movement. For thescope of this work, we drive our FE models kinematically. It is possible to usethe inverse modeling capability of ArtiSynth to estimate the virtual muscleactivation from the kinematics of our implemented FE tongue for a normaland different impaired swallow. This can give an enhanced understating ofthe swallowing physiology from a neurological perspective and aid dysphagiadiagnosis.To conclude, this dissertation has a presented a new frame work for gen-erating biomechanical models by building surface geometries and extractingkinematics from animation models. We have applied our modeling frame97work to generate a 3D oropharyngeal model and simulated a fluid bolusdriven by the model kinematics. We have applied 3D perspective visualiza-tion afforded by our model and are working toward extending the trainingmaterials for standardized dysphagia diagnosis. This work sets up the stagefor enhancing the practice of medical pedagogy using biomechanical modelfor dysphagia diagnosis. The research has been carried out within an inter-disciplinary team of clinicians and engineers across the world, and has leadto a number of ongoing collaborations and projects.98Appendix AMBSImPTM c© scoringguidelineA.1 MBSImPTM c© Components, Scores,and Score Definitions99The MODIFIED BARIUM SWALLOW IMPAIRMENT PROFILE: MBSImP ™ ©Components, Scores, and Score DefinitionsORAL ImpairmentComponent 1—Lip Closure0 = No labial escape1 = Interlabial escape; no progression to anterior lip2 = Escape from interlabial space or lateral juncture; no extension     beyond vermilion border3 = Escape progressing to mid-chin4 = Escape beyond mid-chinComponent 2—Tongue Control During Bolus Hold0 = Cohesive bolus between tongue to palatal seal1 = Escape to lateral buccal cavity/floor of mouth (FOM)2 = Posterior escape of less than half of bolus 3 = Posterior escape of greater than half of bolus Component 3—Bolus Preparation/Mastication0 = Timely and efficient chewing and mashing1 = Slow prolonged chewing/mashing with complete re-collection2 = Disorganized chewing/mashing with solid pieces of bolus          unchewed 3 = Minimal chewing/mashing with majority of bolus unchewedComponent 4—Bolus Transport/Lingual Motion0 = Brisk tongue motion 1 = Delayed initiation of tongue motion2 = Slowed tongue motion3 = Repetitive/disorganized tongue motion  4 = Minimal to no tongue motion       _________________________________________________________      PHARYNGEAL ImpairmentComponent 7—Soft Palate Elevation0 = No bolus between soft palate (SP)/pharyngeal wall (PW)1 = Trace column of contrast or air between SP and PW2 = Escape to nasopharynx3 = Escape to nasal cavity4 = Escape to nostril with/without emissionComponent 8—Laryngeal Elevation0 = Complete superior movement of thyroid cartilage with     complete approximation of  arytenoids to epiglottic petiole1 = Partial superior movement of thyroid cartilage/partial                      approximation of arytenoids to epiglottic petiole2 = Minimal superior movement of thyroid cartilage with minimal       approximation of arytenoids to epiglottic petiole3 = No superior movement of thyroid cartilageComponent 9—Anterior Hyoid Excursion  0 = Complete anterior movement   1 = Partial anterior movement      2 = No anterior movementComponent 10–Epiglottic Movement   0 = Complete inversion   1 = Partial inversion    2 = No inversionComponent 11—Laryngeal Vestibular Closure – Height of Swallow  0 = Complete; no air/contrast in laryngeal vestibule  1 = Incomplete; narrow column air/contrast in laryngeal vestibule  2 = None; wide column air/contrast in laryngeal vestibule Component 12—Pharyngeal Stripping Wave  0 = Present - complete  1 = Present - diminished  2 = Absent               Component 5 – Oral Residue        0 =  Complete oral clearance        1 = Trace residue lining oral structures        2 = Residue collection on oral structures        3 = Majority of bolus remaining        4 = Minimal to no clearance           Location                                         A = Floor of mouth (FOM)               B = Palate                                       C = TongueD = Lateral sulciComponent  6—Initiation of Pharyngeal Swallow0 = Bolus head at posterior angle of ramus (first hyoid excursion)1 = Bolus head in valleculae2 = Bolus head at posterior laryngeal surface of epiglottis3 = Bolus head in pyriforms4 = No visible initiation at any location                 ______________________________________________________________ Component 13—Pharyngeal Contraction (  A/P VIEW ONLY  ) 0 = Complete1 = Incomplete (Pseudodiverticulae)2 = Unilateral Bulging3 = Bilateral BulgingComponent 14—Pharyngoesophageal Segment Opening0 = Complete distension and complete duration; no obstruction of flow1 = Partial distension/partial duration; partial obstruction of flow2 = Minimal distension/minimal duration; marked obstruction of flow3 = No distension with total obstruction of flowComponent 15  —  Tongue Base (TB) Retraction  0 = No contrast between TB and posterior pharyngeal wall (PW)1 = Trace column of contrast or air between TB and PW2 = Narrow column of contrast or air between TB and PW3 = Wide column of contrast or air between TB and PW4 = No visible posterior motion of TBComponent 16  —  Pharyngeal Residue  0 = Complete pharyngeal clearance1 = Trace residue within or on pharyngeal structures2 = Collection of residue within or on pharyngeal structures3 = Majority of contrast within or on pharyngeal structures4 = Minimal to no pharyngeal clearance       Location                A = Tongue Base  B = Valleculae  C = Pharyngeal wall  D = Aryepiglottic folds  E = Pyriform sinuses   F = Diffuse (>3 areas)     __________________________________________________________        ESOPHAGEAL ImpairmentComponent 17—Esophageal Clearance Upright Position0 = Complete clearance; esophageal coating  1 = Esophageal retention      2 = Esophageal retention with retrograde flow below pharyngo-            esophageal segment (PES)3 = Esophageal retention with retrograde flow through PES 4 = Minimal to no esophageal clearanceMBSImP-CSD-070811100Appendix BFormulation for SPH fluidsimulationThis appendix discusses the basics of smoothed particle hydrodynamics (SPH)formulation, implementation will be discussed in details. The Navier–Stokesequations for a Lagrangian system are solved using methods described in[115] for simulating fluids using SPH methods. The SPH formulation is of-ten divided into two major steps in literature [115]. The first step is integralrepresentation and the latter one is particle approximation. The first stepconsists of integral representation of the function and the derivatives of thefunction. The integral representation of the function is further approximatedby adding the values of the neighbouring particles. This makes the particleapproximation of the function at discrete particle level whcih is the otherstep. These steps have been called as kernel estimate or particle estimate inthe literature. In this Appendix kernel approximation and particle approxi-mation will be used throughout to avoid confusion.B.1 Integral representation of a functionThe identity of the integral function f(x) used in SPH formulation is repre-sented as the following equation.f(x) =∫Ωf(x′)∂(x-x′)dx′ (B.1)where f is a function of the position vector x in three dimension. Ω is thevolume of the integral containing x. The δ(x-x′) term is the Dirac deltafunction given byδ(x− x′) ={1, x = x′0, otherwise(B.2)101B.1. Integral representation of a functionThe Delta function makes the integral representation in Equation B.1exact as long as the f(x) is continuous in Ω. Now if the Delta functionis replaced by a so called smoothing kernel function or smoothing functionW (x− x′, h), equation (B.1) becomesf(x) =∫Ωf(x′)W (x− x′, h)dx′ (B.3)In the smoothing kernel function W , h represents the smoothing lengthdefining the neighbourhood area or the area of influence of the function. AsW is not a Dirac function rather an approximation, the integral representa-tion in equation (B.3) called is termed as kernel approximation. In the SPHconvention, the kernel operators is denoted by angle brackets〈〉(see [115]).Equation B.3 can be rewritten following the convention as〈f(x)〉=∫Ωf(x′)W (x− x′, h)dx′ (B.4)Some major properties or conditions of the smoothing function W aredescribed in the following discussion.1. Unity The first one is the unity condition or the normalization condi-tion which simply state that the integration of the smoothing functionprovides unity (Equation B.5).∫Ωf(x′)W (x− x′, h)dx′ = 1 (B.5)2. Delta function property The second condition (Equation B.6) ensuresthat if the smoothing length approaches zero, the smoothing functionshould satisfy the Dirac delta function condition.limh→0W (x− x′, h) = δ(x-x′) (B.6)3. Compact support The dimension of the compact support is defined bythe smoothing length h and a scaling factor κW (x− x′, h) = 0, when | x− x′ |> κh (B.7)102B.1. Integral representation of a functionhere κ is a constant related to the smoothing function for point x thatdefines the smoothing area having non-zero values. The effective area iscalled the support domain of that point. This compact condition makesthe integration domain Ω same as the support domain by localizingthe integration over the entire problem domain. In other words thisproperty transforms a SPH approximation from global operation intoa localized operation. This leads to a set of sparse discretized systemmatrices which reduces the computational burden significantly.4. Positivity At any point x′ inside the support domain of x the smooth-ing function has a positive value, that is W (x − x′) ≥ 0. This is aimportant requirement to ensure a physically meaningful representa-tion. For example, In hydrodynamic simulations negative value for thesmoothing function can result in negative density and energy termswhich leads to misrepresentation of the system.5. Decay The value of the smoothing function should decrease with theincrease of the distance away from the particle. In other words what itmeans is nearer particle should have a bigger influence on the concernedparticle.6. Symmetry property The smoothing function should be an even function.What it means is that particles at same distance should have equaleffect on a given particle. However, this is not strictly enforced, and itis sometimes violated to provide higher consistency.7. Smoothness The smoothing function should be sufficiently smooth.This so-called “smoothness” means that the kernel should be suffi-ciently continuous and smooth in order to insensitive to particle disor-der and the errors in approximating the integral.Any function having the above properties can be used as a SPH smoothingfunction and in literature many researchers have proposed different kind tokernel in order to achieve better approximation and computational efficiency.Some of such proposed kernels will be discussed in Section B.4.The error of the SPH integral representation can be estimated roughlyby expanding the smoothing function | x− x′ |> κh from Equation B.7 andsubstituting the value in Equation B.4 which yields103B.2. Integral representation of the derivative of a function〈f(x)〉=∫Ω[f(x) + f ′(x)(f(x′)− x) + r((x′ − x)2)W (x− x′, h)= f(x)∫ΩW (x− x′, h)+ f(x′)∫Ω(x− x′)W (x− x′, h)dx′ + r(h2)(B.8)here r refers to the residual term. Now, the smoothing function W chosen tobe a even function with respect to x, therefore (x − x′)W (x − x′, h) shouldbe an odd function which will mean that∫Ω(x− x′)W (x− x′, h)dx′ = 0 (B.9)Using the normalization condition (Equation B.5) and Equation B.9 thekernel approximation operator derived in Equation B.8 becomes〈f(x)〉= f(x) + r(h2) (B.10)From the above equation it can be seen the kernel approximation has aresidual term with h2 which means in the SPH method the kernel approxi-mation has second order accuracy. However, if the smoothing function is nota even a function or the normalization condition in not satisfied; the kernelapproximation might not be of second order accuracy (detailed discussion in[115]).B.2 Integral representation of the derivativeof a functionThe integral representation of the derivative of the function is obtained byreplacing f(x) with ∇· f(x) in the Equation B.4 which yields〈∇· f(x)〉 = ∫Ω[∇· f(x′)]W (x− x′, h)dx′ (B.11)104B.2. Integral representation of the derivative of a functionhere the divergence is the integral operator with respect to the coordinates.Further, the term inside the integral operator gives[∇· f(x′)]W (x− x′, h) =∇· [f(x′)W (x− x′, h)]−f(x′)· ∇W (x− x′, h) (B.12)Using the above equation to replace the term inside the integral operator,Equation B.11 becomes〈∇· f(x)〉 = ∫Ω∇· [f(x′)W (x− x′, h)]dx′−∫Ωf(x′)· ∇W (x− x′, h)dx′(B.13)The first integral term of Equation B.13 can be converted to an integralover the whole surface S on the integration domain Ω with a unit vector −→nwhich is normal to the surface S.〈∇· f(x)〉 = ∫Sf(x′)W (x− x′, h)· −→n dS−∫Ωf(x′)· ∇W (x− x′, h)dx′(B.14)The smoothing function for a point W has compact support (conditionthree Equation B.7) when the support domain is located within the prob-lem domain (Figure B.1). For such points the surface integral becomes 0.Considering a point whose support domain intersects the problem domain,the smoothing function W is truncated at the boundary of problem domain(Figure B.2) making the surface integral nonzero. For such cases, modifica-tion should be made to remedy the boundary conditions. Nonetheless, forthe points having a zero surface integral Equation B.14 becomes〈∇· f(x)〉 = −∫Ωf(x′)· ∇W (x− x′, h)dx′ (B.15)The above equation shows that spatial gradient of a field function in SPHrepresentation is calculated from the values of the function and the derivative105B.2. Integral representation of the derivative of a functionFigure B.1: The support domain of the smoothing function W and problemdomain. The support domain is located within the problem domain. There-fore, the surface integral on the right hand side of Equation B.14 is zero.Adapted from [115].Figure B.2: The support domain of the smoothing function W and problemdomain. The support domain intersects with the problem domain. Therefore,the smoothing function W is truncated by the boundary, and the surfaceintegral on the right hand side of Equation B.14 is on longer zero. Adaptedfrom [115].of the smoothing function W , not the derivative of the function itself. Thisis significant because it reduces the consistency requirement on the assumedfield functions for getting stable solution for the PDEs which is very similarto the weak form methods [123].106B.3. Particle approximationB.3 Particle approximationAfter the kernel approximation (discussed in Section B.1 and B.2), anotherkey operation in the SPH methodology is particle approximation. In thisstep the SPH kernel derived in Equation B.4 and B.15 are translated todiscretized forms of summation over all the particles in the support domain.A particle j(= 1, 2, . . . , N) with a support domain that has N number ofparticles, is chosen with mass mj and density ρj. If the infinitesimal volumedx′ in the integration function for particle j is replaced with finite volumeof the particle ∇Vj, the mass can be written as mj = ∇Vjρj. With theseapproximations in the continuous integral representation of f(x) becomes thediscretized particle approximation explained in the following derivation.f(x) =∫Ωf(x′)W (x− x′, h)dx′≡N∑j=1f(xj)W (x− xj, h)∇Vj=N∑j=1f(xj)W (x− xj, h) 1ρj(ρj∇Vj)=N∑j=1mjρjf(x′)W (x− xj, h)(B.16)The particle approximation for a function at particle i can be written asthe following 〈f(xi)〉=N∑j=1mjρjf(xj)Wij (B.17)whereWij = W (xi − xj, h) = W (|xi − xj|, h) (B.18)The particle approximation represented by Equation B.17 implies that thevalue of a function at particular particle is approximated using the averageof that functions at all the particles in the support domain weighted by thesmoothing function.107B.3. Particle approximationTo use the particle approximation of the spatial derivative of the function,same assumptions like Equation B.16 can be made to achieve the followingequation where the gradient of the smoothing function W is taken with re-spect to the particle j.〈∇· f(x)〉 = − N∑j=1mjρjf(xj)· ∇W (x− xj, h) (B.19)Now the particle approximation for a function at particle i, denoting thedistance between particle i and j as rij, can be written as〈∇· f(xi)〉 = − N∑j=1mjρjf(xj)· ∇Wij (B.20)where∇iWij = xi − xjrij∂Wij∂rij=xijrij∂Wij∂rij(B.21)As ∇iWij is taken with respect to particle i, the negative sign in EquationB.18 is removed in the following equation.〈∇· f(xi)〉 = N∑j=1mjρjf(xj)· ∇Wij (B.22)This particle approximation of the derivative of a function stated that thevalue of the gradient of the function at a particular particle is approximatedby averaging the value of the gradient of the function on all the particlesin the neighbourhood or the support domain, weighted by the gradient ofthe smoothing function. The use of particle summations to approximatethe integral makes the SPH method a meshfree method without the needof a background mesh for numerical integration. The particle approxima-tion (Equation B.21) includes the mass and density of the particle in theformulation. This makes the SPH formulation convenient for hydrodynamicapplication because density is one of the key field variable for dynamic fluidflow problems. By substituting the the function f(x) in Equation B.17 withdensity ρ the SPH approximation for the density function becomesρi =N∑j=1mjWij (B.23)108B.4. Choice of smoothing functionwhich is referred as summation density approach which is commonly usedto obtain density in a SPH simulation. This equation states that the densityof a particle is weighted average of all the neighbouring particles inValue of a function Value of its derivatives〈f(xi)〉=N∑j=1mjρjf(xj)·Wij〈∇· f(xi)〉 =N∑j=1mjρjf(xj)· ∇Wijwhere whereWij = W (xi − xj, h)= W (|xi − xj|, h)∇iWij= xi − xjrij∂Wij∂rij=xijrij∂Wij∂rijTable B.1: Particle approximation representation of the value of a functionand its derivatives at particle iParticle approximation representation of a field function and its deriva-tives for a given particle i can be summarized as in the following Table B.1,which explains the fundamental advantage of the SPH method; that is thederivatives of any function f(x) can be found by differentiating the kernelrather than by using finite difference, finite volume or finite element expres-sion calculated using the grid.B.4 Choice of smoothing functionIn SPH formulation, the choice of smoothing function is very important be-cause it determines how effectively functional approximation is performedbased on scattered nodes in the support domain without using a predefinedgrid that provides the connectivity of the nodes. This smoothing functionis referred as smoothing kernel function, smoothing kernel or simply kernelin literature. Many researchers have proposed different kernel functions foroptimal performance.In the original paper Lucy [102] proposed a bell shaped curve as described109B.5. SPH equation of motionby the following equation as the smoothing function.W (x− x′, h) = W (R, h)= αd{(1 + 3R)(1−R)3 R ≤ 10 R > 1(B.24)here R is the relative distance between two particles located at x and x′,yielding R =rh=|x− x′|h. αd satisfies the unity condition in all threedimensions having values 5/4h, 5/pih2 and 105/16pih3 in one-,two- and threedimension respectively. Monaghan [101] in his seminal paper suggested toassume the smoothing function to be a Gaussian for simulating non-sphericalstars.W (R, h) = αde−R2 (B.25)where αd is 1/pi1/2h, 1/pih2 and 1/pi3/2h3 in one-,two- and three- dimentionalspace respectively.Later on Monaghan and Lattamzio [124] defined a smoothing functionbased on the cubic splines popularly know as B-spline function. This func-tion has been most widely used. Morris [125] introduced quartic splines tomore closely approximate the Gaussian ensuring more stability. Johnson etal. [126] used a quadratic smoothing function as it could overcome the com-pressive instability problem which occurs in high velocity impact problems.More details about these smoothing functions can be found at Chapter 3 of[115].B.5 SPH equation of motionIn the previous sections essential formulations and the smoothing functionsfor the SPH methods has been discussed for discretizing the PDEs. In thissection some further modification in the formulation for these numerical pro-cedures are made to facilitate dynamic fluid simulation. As discussed in2.4.4, SPH describes the physical governing equations using an Lagrangianapproach. In this section the SPH equations of motion will be derived basedon such governing equations in Lagrangian form.In a Lagrangian description, this control volume V will move with thefluid flow given that the mass of the fluids contained in V remains unchanged.Considering an infinitesimal fluid cell with control volume δV bounded by110B.5. SPH equation of motioncontrol surface δS, which can be the differential volume dV and the differ-ential surface dS. The conditions for selecting such infinitesimal fluid cellare (1) the cell is large enough that the continuum mechanics assumptionsare valid and (2) the cell is small enough that fluid properties inside thecell can be regarded isotropic. This infinitesimal fluid cell can move along astreamline with vector velocity v = vx, vy, vz. In case of a Lagrangian controlvolume, the movement of fluids inside the control volume V changes the con-trol surface S which again changes the control volume. So the change in thecontrol volume ∆V due to the movement of dS over a Small time increment∆t can be written as∆V = v∆t·ndS (B.26)where n is the unit vector perpendicular to the surface dS.The total volume change can be therefore calculated by integrating overthe control surface S yielding∆V =∫Sv∆t·ndS (B.27)Applying the divergence theorem with a gradient operator ∇ after divid-ing the both sides of Equation B.27 by ∆t, it becomes∆V∆t=∫V∇·vdV (B.28)One of the condition of selecting a small enough infinitesimal fluid cellwas that the fluid properties would be isotropic throughout. What it impliesthat if the entire control volume V is was considered as a small as the fluidcell volume δV , the fluid property would have been isotropic throughout δV .The following equation can be obtained with such assumptions∆(δV )∆t= (∇·v)∫Vd(δV ) = (∇·v)δV (B.29)Hence, the time rate of volume change for the infinitesimal fluid cell isfound to beD(δV )Dt= (∇·v)δV (B.30)111B.6. The continuity equationRearranging the above equation the velocity divergence is found to be∇·v = 1δVD(δV )Dt(B.31)The above equation physically interprets the velocity divergence as the timerate of volume change per unit volume.Three fundamental conservation law govern the basic fluid dynamicsequation. They are1. Conservation of mass or the continuity equation2. Conservation of momentum or the momentum equation3. Conservation of energy or the energy equationIn the following sections, these governing equations in the Lagrangianform will be discussed for SPH method.B.6 The continuity equationThe continuity equation is the law of conservation of mass. For a fluid cellof volume δV , the mass contained in the control volume isδm = ρδV (B.32)where m is the density and ρ is the density. According to the conservationof mass law, no mass can be created or destroyed. So the mass is conservedin the fluid cell and time rate of mass change is zero.D(δm)Dt=D(ρδV )Dt= 0= δVDρDt+ ρD(δV )Dt= 0(B.33)which can be rewritten asDρDt+ ρ1δVD(δv)Dt= 0 (B.34)From the physical interpretation of velocity divergence (Equation B.31)and the above equation, the continuity equation in Lagrangian form can beobtained asDρDt= −ρ∆·v (B.35)112B.7. The momentum equationB.7 The momentum equationThe momentum equation is represented by Newton’s second law. For aninfinitesimal Lagrangian fluid cell, the net force acting on the cell is equal tothe mass contained in the cell times acceleration if that fluid cell. The netforce acting on the fluid cell includes body forces i.e. gravitational, magneticand surface forces i.e. pressure, shear and normal stress. The surface forcedue to pressure p is imposed by the surrounding fluid cells and the shear andnormal stress is the outcome of deformations and volume change respectively.The acceleration of the fluid cell with position vector x = (x, y, z) in threedirection are Dvx/Dt, Dvy/Dt and Dvz/Dt. Considering τij as the stress inj direction acting on a plane perpendicular to i axis and the body force perunit mass in x direction is Fx, Newton’s second law can be written asmdvxdt=ρdxdydzdvxdt=− ∂p∂xdxdydz+∂τxx∂xdxdydz +∂τyx∂ydxdydz +∂τzx∂zdxdydz+ Fx(ρdxdydz)(B.36)The equation of momentum in x direction becomesρDvxDt= −∂p∂x+∂τxx∂x+∂τyx∂y+∂τzx∂z+ ρFx (B.37)Momentum equation in y and z direction have similar form like the aboveequation.B.8 The energy equationThe energy equation is based on the first law of thermodynamics that isconservation of energy. The energy equation states that the time rate ofenergy change inside an infinitesimal fluid cell is the summation of (1) thework done by pressure multiplied by the volumetric strain and (2) the energy113B.9. Navier-Stokes equationsdissipation due to the viscous shear forces.ρDeDt= p(∂vx∂x+∂vy∂y+∂vz∂z)+ τxx∂vx∂x+ τyx∂vx∂y+ τzx∂vx∂z+ τxy∂vy∂x+ τyy∂vy∂y+ τzy∂vy∂z+ τxz∂vz∂x+ τyz∂vz∂y+ τzz∂vx∂z(B.38)B.9 Navier-Stokes equationsNavier-Stokes equations are a set of partial differential equation using a La-grangian description which states the conservation of mass, momentum andenergy to govern dynamic fluid flow. The Navier-Stokes equations consistsof the following set of equations.1. The Continuity equationDρDt= −ρ∂vβ∂xβ(B.39)2. The momentum equationDvαDt=1ρ∂σαβ∂xβ(B.40)3. The energy equationDeDt=σαβρ∂vα∂xβ(B.41)In the above equations α and β are used to denote the coordinate systemwhere the summation in the equations are taken over repeated indices. σ isthe total stress tensor which is expressed asσαβ = −pδαβ + ταβ (B.42)here p is the isotropic pressure and τ is the viscous stress.114B.10. Particle approximation of densityB.10 Particle approximation of densityIn SPH method density determines the particle distribution and smoothinglength evaluation. There are two major approaches to evolve density usingSPH mehtod. The first approach is summation density where SPH approxi-mation is directly applied to density itself. Following such approach, densityρ for a particle i can be written in the form ofρi =N∑j=1mjWij (B.43)where N number of particles reside in the support domain of i, mj is themass associate with particle j and Wij is the smoothing function of particlei evaluated at particle j which has an unit of inverse of volume. So thisequation states that the density of a particle is approximated by averagingthe densities of the particles in the support domain.There is another approach for approximating the density which is calledcontinuity density approach. If SPH approximation is applied to only to thevelocity divergence part in the continuity equation (B.39), it becomesDρiDt= −ρiN∑j=1mjρjvβj ·∂Wij∂bβi(B.44)Density can also be approximated using continuity density by applyingthe following identity to place the density term inside the gradient operator− ρ∂vβ∂xβ= −(∂(ρvβ)∂xβ− vβ· ∂ρ∂xβ)(B.45)If the SPH approximation is only applied at every gradient and to thevelocity at the second term in the right hand side of the Equation B.46, thecontinuity density equation becomesDρiDt=N∑j=1mjvβij·∂Wij∂xβi(B.46)where vβij =(vβi − vβj)which introduces velocity difference into the discreteparticle approximation. Equation B.46 shows that the time rate of density115B.11. Particle approximation of momentumchange of particle is dependent on relative velocity between the particle andthe other particles in the support domain weighted by the gradient of thesmoothing function.There are comparative advantages and disadvantages of both approachwhich depended on the application. General fluid continuity problems with-out discontinuities like shock-waves can be simulated with the simulationdensity approach. On other hand for simulating phenomena with strongdiscontinuity e.g. explosion, high velocity impact, the continuity density ap-proach produces better results.B.11 Particle approximation of momentumMomentum can be approximated using SPH formulation like density was inthe last section. This approximation is achieved by directly applying SPHparticle approximation concepts to the gradients of the momentum equation(Equation B.40).DvαiDt=1ρiN∑j=1mjσαβjρj∂Wij∂xβi(B.47)By defining different identities different formulations for approximatingmomentum can be achieved which over various advantages for simulatingdifferent type of problems. One frequently used formulation of momentumdefines the following identityN∑j=1mjσαβiρiρj∂Wij∂xβi=σαβiρi( N∑j=1mjρj∂Wij∂xβi)(B.48)which yieldsDvαiDt=N∑j=1mjσαβi + σαβjρiρj∂Wij∂xji(B.49)The above formulation takes advantage of the summation representationwhich in term reduces error arising from the particle inconsistency problem(Chapter 3 of [115]).116B.12. Particle approximation of energyB.12 Particle approximation of energyFor Newtonian fluids the viscous shear stress τ is proportional to the shearstrain ε with dynamic viscosity µ.ταβ = µεαβ (B.50)whereεαβ =∂vβ∂xα+∂vα∂xβ− 23(∇·v)δαβ (B.51)Separating isotropic pressure and the viscous stress in energy equation(Equation B.41), it becomesDeDt= −pρ∂vβ∂xβ+µ2ρεαβεαβ (B.52)To evaluate the internal energy e in the above equation, the pressurework is derived following SPH formulations. The continuity density equation(Equation B.46) at particle i can be approximated as− pρ∂vβi∂xβi=piρ2iN∑j=1mjvβij·∂Wij∂xβi(B.53)Considering the following identity− pρ∂vβ∂xβ= − ∂∂xβ(pvβρ)+ vβ∂∂xβ(pρ)(B.54)if considered for particle i, using SPH approximation the above identity be-comes− pρ∂vβi∂xβi=N∑j=1mjpjρ2jvβij·∂Wij∂xβi(B.55)summing Equation B.53 and B.55 yields− pρ∂vβi∂xβi=12N∑j=1mj(piρ2i+pjρ2j)vβij·∂Wij∂xβi(B.56)So the SPH formulation for internal energy evaluation (Equation B.52)for a particle i with N number of particles in its support domain becomesDeiDt=12N∑j=1mj(piρ2i+pjρ2j)vβij∂Wij∂xβi+µi2ρiεαβi εαβi (B.57)117Bibliography[1] K. Matsuo and J. B. Palmer, “Anatomy and physiology of feeding andswallowing: normal and abnormal,” Physical medicine and rehabilita-tion clinics of North America, vol. 19, no. 4, pp. 691–707, 2008. →page 1[2] B. Bloem, A. Lagaay, W. Van Beek, J. Haan, R. Roos, and A. Wintzen,“Prevalence of subjective dysphagia in community residents aged over87.” Bmj, vol. 300, no. 6726, pp. 721–722, 1990. → page 1, 2[3] S. Lindgren and L. Janzon, “Prevalence of swallowing complaints andclinical findings among 50–79-year-old men and women in an urbanpopulation,” Dysphagia, vol. 6, no. 4, pp. 187–192, 1991. → page 1, 2[4] G. Pugliese and D. A. Lichtenberg, “Nosocomial bacterial pneumonia:an overview,” American journal of infection control, vol. 15, no. 6, pp.249–265, 1987. → page 1[5] C. S. Bryan and K. L. Reynolds, “Bacteremic nosocomial pneumonia:Analysis of 172 episodes from a single metropolitan area 1, 2,” Ameri-can Review of Respiratory Disease, vol. 129, no. 5, pp. 668–671, 1984.→ page 1[6] R. Wenzel, “Hospital-acquired pneumonia: overview of the currentstate of the art for prevention and control,” European Journal of Clini-cal Microbiology and Infectious Diseases, vol. 8, no. 1, pp. 56–60, 1989.→ page 1[7] T. J. Marrie, H. Durant, and C. Kwan, “Nursing home-acquired pneu-monia: A case-control study,” Journal of the American Geriatrics So-ciety, vol. 34, no. 10, pp. 697–702, 1986. → page 1[8] O. Ekberg, S. Hamdy, V. Woisard, A. Wuttge-Hannig, and P. Ortega,“Social and psychological burden of dysphagia: its impact on diagnosis118Bibliographyand treatment,” Dysphagia, vol. 17, no. 2, pp. 139–146, 2002. → page1[9] J. Penman and M. Thomson, “A review of the textured diets developedfor the management of dysphagia,” Journal of Human Nutrition andDietetics, vol. 11, no. 1, pp. 51–60, 1998. → page 2[10] Y. Noh, M. Segawa, K. Sato, C. Wang, H. Ishii, J. Solis, A. Takanishi,A. Katsumata, and Y. Iida, “Development of a robot which can sim-ulate swallowing of food boluses with various properties for the studyof rehabilitation of swallowing disorders,” in Robotics and automation(ICRA), 2011 IEEE international conference on. IEEE, 2011, pp.4676–4681. → page 3, 30[11] J. Remmers, E. Sauerland, A. Anch et al., “Pathogenesis of upperairway occlusion during sleep,” Journal of Applied Physiology, vol. 44,no. 6, pp. 931–938, 1978. → page 3[12] D. Friberg, T. Ansved, K. Borg, B. Carlsson-Nordlander, H. LARS-SON, and E. Svanborg, “Histological indications of a progressive snor-ers disease in an upper airway muscle,” American journal of respiratoryand critical care medicine, vol. 157, no. 2, pp. 586–593, 1998. → page3[13] R. JOHN KIMOFF, E. Sforza, V. Champagne, L. Ofiara, and D. Gen-dron, “Upper airway sensation in snoring and obstructive sleep apnea,”American journal of respiratory and critical care medicine, vol. 164,no. 2, pp. 250–255, 2001. → page 3[14] D. Friberg, B. Gazelius, T. Ho¨kfelt, and B. Nordlander, “Abnormalafferent nerve endings in the soft palatal mucosa of sleep apnoics andhabitual snorers,” Regulatory peptides, vol. 71, no. 1, pp. 29–36, 1997.→ page 3[15] “Omnibus survey: caseloads for speech-language pathologists, amer-ican speech-language-hearing association and others,” rockville, MD:ASHA, 2000. → page 4[16] T. M. Kidder, S. E. Langmore, and B. J. Martin, “Indications and tech-niques of endoscopy in evaluation of cervical dysphagia: comparison119Bibliographywith radiographic techniques,” Dysphagia, vol. 9, no. 4, pp. 256–261,1994. → page 4[17] B. Martin-Harris, M. B. Brodsky, Y. Michel, D. O. Castell, M. Schle-icher, J. Sandidge, R. Maxwell, and J. Blair, “MBS measurement toolfor swallow impairment–MBSImP: establishing a standard,” Dyspha-gia, vol. 23, no. 4, pp. 392–405, 2008. → page 4, 15, 23, 31[18] J. A. Logemann and J. A. Logemann, “Evaluation and treatment ofswallowing disorders,” 1983. → page 5, 72[19] B. Martin-Harris, Y. Michel, and D. O. Castell, “Physiologic modelof oropharyngeal swallowing revisited,” Otolaryngology-Head and NeckSurgery, vol. 133, no. 2, pp. 234–240, 2005. → page 5[20] I. J. Cook, W. J. Dodds, R. O. Dantas, M. K. Kern, B. T. Massey,R. Shaker, and W. J. Hogan, “Timing of videofluoroscopic, manometricevents, and bolus transit during the oral and pharyngeal phases ofswallowing,” Dysphagia, vol. 4, no. 1, pp. 8–15, 1989. → page 5[21] F. McConnel, D. Cerenko, and M. Mendelsohn, “Manofluorographicanalysis of swallowing.” Otolaryngologic Clinics of North America,vol. 21, no. 4, pp. 625–635, 1988. → page 5[22] R. Olsson, J. Castell, D. Castell, and O. Ekberg, “Solid-state comput-erized manometry improves diagnostic yield in pharyngeal dysphagia:simultaneous videoradiography and manometry in dysphagia patientswith normal barium swallows,” Abdominal imaging, vol. 20, no. 3, pp.230–235, 1995. → page 5[23] W. J. Dodds, “The physiology of swallowing,” Dysphagia, vol. 3, no. 4,pp. 171–178, 1989. → page 6[24] S. M. Shaw and R. Martino, “The normal swallow: muscular andneurophysiological control,” Otolaryngologic clinics of North America,vol. 46, no. 6, pp. 937–956, 2013. → page 8[25] E. R. Blonsky, J. A. Logemann, B. Boshes, and H. B. Fisher, “Compari-son of speech and swallowing function in patients with tremor disordersand in normal geriatric patients: a cinefluorographic study,” Journalof Gerontology, vol. 30, no. 3, pp. 299–303, 1975. → page 8, 10120Bibliography[26] ““hyoglossus” by image:gray1019.png modiefied by uwe gille viawikimedia commons,”,accessed: 2015-07-14. → page 8[27] ““mouth illustration–otis archives” by duncan kenneth win-ter via wikimedia commons,” illustration-Otis Archives.jpg#/media/File:Mouth illustration-Otis Archives.jpg, accessed: 2015-07-14. → page 8[28] W. Pommerenke, “A study of the sensory areas eliciting the swallowingreflex,” American Journal of Physiology–Legacy Content, vol. 84, no. 1,pp. 36–41, 1928. → page 9[29] J. Robbins, G. Gensler, J. Hind, J. Logemann, A. Lindblad, D. Brandt,H. Baum, D. Lilienfeld, S. Kosek, D. Lundy et al., “Can thickenedliquids or chin-down posture prevent aspiration?” 2008. → page 9[30] J. A. Logemann, B. R. Pauloski, L. Colangelo, C. Lazarus, M. Fujiu,and P. J. Kahrilas, “Effects of a sour bolus on oropharyngeal swallowingmeasures in patients with neurogenic dysphagia,” Journal of Speech,Language, and Hearing Research, vol. 38, no. 3, pp. 556–563, 1995. →page 9[31] B. Martin-Harris, M. B. Brodsky, Y. Michel, F.-S. Lee, and B. Walters,“Delayed initiation of the pharyngeal swallow: normal variability inadult swallows,” Journal of Speech, Language, and Hearing Research,vol. 50, no. 3, pp. 585–594, 2007. → page 9[32] W. J. Dodds, E. T. Stewart, and J. A. Logemann, “Physiology andradiology of the normal oral and pharyngeal phases of swallowing.”AJR. American journal of roentgenology, vol. 154, no. 5, pp. 953–963,1990. → page 9, 10[33] F. Mcconnel, “Analysis of pressure generation and bolus transit duringpharyngeal swallowing,” The Laryngoscope, vol. 98, no. 1, pp. 71–78,1988. → page 10[34] W. G. Pearson Jr, S. E. Langmore, B. Y. Louis, and A. C. Zumwalt,“Structural analysis of muscles elevating the hyolaryngeal complex,”Dysphagia, vol. 27, no. 4, pp. 445–451, 2012. → page 10121Bibliography[35] “Human physiology: Digestion,”, accessed: 2015-07-14. → page 11[36] T. J. Humphries and D. O. Castell, “Pressure profile of esophagealperistalsis in normal humans as measured by direct intraesophagealtransducers,” The American journal of digestive diseases, vol. 22, no. 7,pp. 641–645, 1977. → page 11[37] R. K. Goyal and B. W. Cobb, “Motility of the pharynx, esophagus andesophageal sphincters,” Physiology of the gastrointestinal tract, vol. 1,pp. 359–391, 1981. → page 11[38] N. De Vincentis, R. Lenti, C. Pona, A. Lococo, R. Crisci, G. Coloni,I. Flutsch, and G. Mariani, “Scintigraphic evaluation of the esophagealtransit time for the noninvasive assessment of esophageal motor dis-orders,” J Nucl Med Allied Sci, vol. 28, pp. 137–142, 1984. → page12[39] B. Martin, J. Logemann, R. Shaker, and W. Dodds, “Coordinationbetween respiration and swallowing: respiratory phase relationshipsand temporal integration,” Journal of Applied Physiology, vol. 76, no. 2,pp. 714–723, 1994. → page 12[40] M. S. Klahn and A. L. Perlman, “Temporal and durational patternsassociating respiration and swallowing,” Dysphagia, vol. 14, no. 3, pp.131–138, 1999. → page 12[41] S. Jafari, R. A. Prince, D. Y. Kim, and D. Paydarfar, “Sensory reg-ulation of swallowing and airway protection: a role for the internalsuperior laryngeal nerve in humans,” The Journal of physiology, vol.550, no. 1, pp. 287–304, 2003. → page 12[42] N. R. BLEACH, “The gag reflex and aspiration: a retrospective anal-ysis of 120 patients assessed by videofluoroscopy,” Clinical Otolaryn-gology & Allied Sciences, vol. 18, no. 4, pp. 303–307, 1993. → page12[43] G. Mann and G. J. Hankey, “Initial clinical and demographic predictorsof swallowing impairment following acute stroke,” Dysphagia, vol. 16,no. 3, pp. 208–215, 2001. → page 13122Bibliography[44] A. R. Maccarini, A. Filippini, D. Padovani, M. Limarzi, M. Loffredo,and D. Casolino, “Clinical non-instrumental evaluation of dysphagia,”Acta otorhinolaryngologica italica, vol. 27, pp. 299–305, 2007. → page14[45] P. Linden and A. Siebens, “Dysphagia: predicting laryngeal penetra-tion.” Archives of Physical Medicine and Rehabilitation, vol. 64, no. 6,pp. 281–284, 1983. → page 14[46] “Fluoroscope,”, accessed: 2015-07-14. → page 14[47] J. B. Palmer, K. V. Kuhlemeier, D. C. Tippett, and C. Lynch, “A pro-tocol for the videofluorographic swallowing study,” Dysphagia, vol. 8,no. 3, pp. 209–214, 1993. → page 15, 21[48] K. Kuhlemeier, P. Yates, and J. Palmer, “Intra-and interrater variationin the evaluation of videofluorographic swallowing studies,” Dysphagia,vol. 13, no. 3, pp. 142–147, 1998. → page 15, 21[49] B. Martin-Harris, J. A. Logemann, S. McMahon, M. Schleicher, andJ. Sandidge, “Clinical utility of the modified barium swallow,” Dys-phagia, vol. 15, no. 3, pp. 136–141, 2000. → page 16[50] G. Ramsey, J. Watson, R. Gramiak, and S. Weinberg, “Cinefluoro-graphic analysis of the mechanism of swallowing 1,” Radiology, vol. 64,no. 4, pp. 498–518, 1955. → page 16[51] O. Ekberg and S. V. Sigurjo´nsson, “Movement of the epiglottis duringdeglutition,” Gastrointestinal radiology, vol. 7, no. 1, pp. 101–107, 1982.→ page 16[52] “Fiberoptic Endoscopic Evaluation of Swallowing (FEES): Evaluatethe pharyneageal stage of swallowing and diagnose dysphagia.” scope.html, accessed: 2015-07-14. → page 17[53] S. E. Langmore, S. M. Kenneth, and N. Olsen, “Fiberoptic endoscopicexamination of swallowing safety: a new procedure,” Dysphagia, vol. 2,no. 4, pp. 216–219, 1988. → page 16123Bibliography[54] “View of the glottis as seen during laryngoscopy.” endo 2.jpg, accessed: 2015-07-14. →page 18[55] “Anatomical parts seen during laryngoscopy by alan hoofring.”→ page18[56] S. E. Langmore, “Endoscopic evaluation of oral and pharyngeal phasesof swallowing,” GI Motility online, 2006. → page 18[57] ““achalasia ems” by jc petit,” EMS.jpg#/media/File:Achalasia EMS.jpg, accessed: 2015-07-14. → page 19[58] J. E. Pandolfino and P. J. Kahrilas, “Aga technical review on the clin-ical use of esophageal manometry,” Gastroenterology, vol. 128, no. 1,pp. 209–224, 2005. → page 19[59] “Ultrasound images of the tongue nevit dilmen,” ScanND 110201103913 1047580.jpg, accessed: 2015-07-14. → page 20[60] T. H. Shawker, B. Sonies, M. Stone, and B. J. Baum, “Real-time ultra-sound visualization of tongue movement during swallowing,” Journalof Clinical Ultrasound, vol. 11, no. 9, pp. 485–490, 1983. → page 20[61] T. H. Shawker, B. C. Sonies, and M. Stone, “Soft tissue anatomy of thetongue and floor of the mouth: an ultrasound demonstration,” Brainand language, vol. 21, no. 2, pp. 335–350, 1984. → page 20[62] T. Shawker, B. Sonies, and M. Stone, “Sonography of speech and swal-lowing,” Ultrasound annual, pp. 237–260, 1984. → page 21[63] J. Benson and D. Tuchman, “Other diagnostic tests used for evalua-tion of swallowing disorders,” Disorders of Feeding and Swallowing inInfants and Children: Pathophysiology, Diagnosis and Treatment (1stedn). San Diego: Singular, pp. 201–9, 1994. → page 21[64] B. C. Sonies, G. Chi-Fishman, and J. L. Miller, “Ultrasound imagingand swallowing,” in Normal and Abnormal Swallowing. Springer, 2003,pp. 119–138. → page 21124Bibliography[65] J. A. Cichero and B. E. Murdoch, Dysphagia: foundation, theory andpractice. John Wiley & Sons, 2006. → page 21[66] F. E. Ardakani, “Evaluation of swallowing patterns of the tongue usingreal-time b-mode sonography,” J Contemp Dent Pract, vol. 7, no. 3,pp. 67–74, 2006. → page 21[67] W. Chih-Hsiu, H. Tzu-Yu, C. Jiann-Chyuan, C. Yeun-Chung, andL. Shiann-Yann, “Evaluation of swallowing safety with fiberoptic en-doscope: comparison with videofluoroscopic technique,” The Laryngo-scope, vol. 107, no. 3, pp. 396–401, 1997. → page 21, 22[68] M. Sitton, J. Arvedson, A. Visotcky, N. Braun, J. Kerschner, S. Tarima,and D. Brown, “Fiberoptic endoscopic evaluation of swallowing inchildren: feeding outcomes related to diagnostic groups and endo-scopic findings,” International journal of pediatric otorhinolaryngology,vol. 75, no. 8, pp. 1024–1031, 2011. → page 21, 22[69] S. Pe´rie´, L. Laccourreye, A. Flahault, V. Hazebroucq, S. Chaussade,and J. L. S. Guily, “Role of videoendoscopy in assessment of pharyn-geal function in oropharyngeal dysphagia: comparison with videoflu-oroscopy and manometry,” The Laryngoscope, vol. 108, no. 11, pp.1712–1716, 1998. → page 21, 22[70] J. A. Logemann, A. W. Rademaker, B. R. Pauloski, Y. Ohmae, andP. J. Kahrilas, “Normal swallowing physiology as viewed by videofluo-roscopy and videoendoscopy,” Folia phoniatrica et logopaedica, vol. 50,no. 6, pp. 311–319, 1998. → page 21, 22[71] “Translating research into practice (TRIP)-II. fact sheet. agency forhealth quality research.” → page 22[72] B. Martin-Harris and B. Jones, “The videofluorographic swallowingstudy,” Physical medicine and rehabilitation clinics of North America,vol. 19, no. 4, pp. 769–785, 2008. → page 22, 23[73] G. J. Bours, R. Speyer, J. Lemmens, M. Limburg, and R. De Wit,“Bedside screening tests vs. videofluoroscopy or fibreoptic endoscopicevaluation of swallowing to detect dysphagia in patients with neurologi-cal disorders: systematic review,” Journal of advanced nursing, vol. 65,no. 3, pp. 477–493, 2009. → page 22125Bibliography[74] L. Perry and C. P. Love, “Screening for dysphagia and aspiration inacute stroke: a systematic review,” Dysphagia, vol. 16, no. 1, pp. 7–18,2001. → page 22[75] R. Martino, G. Pron, and N. Diamant, “Screening for oropharyngealdysphagia in stroke: insufficient evidence for guidelines,” Dysphagia,vol. 15, no. 1, pp. 19–30, 2000. → page 22[76] “Modified barium swallow impairment profile. northern speech ser-vices.”, accessed: 2015-07-14. → page 23,24, 31, 77[77] J. Cooper and V. Taqueti, “A brief history of the development ofmannequin simulators for clinical education and training,” Quality andSafety in Health Care, vol. 13, no. suppl 1, pp. i11–i18, 2004. → page29[78] “Resusci Anne via Wikimedia Commons,” Anne.jpg, accessed: 2015-09-22.→ page 29[79] “The newer version of “harvey” mannequin along with it’s basevia wikimedia commons,” new.jpg, accessed: 2015-09-22. → page 29[80] A. Grenvik and J. Schaefer, “From resusci-anne to sim-man: the evo-lution of simulators in medicine,” Critical care medicine, vol. 32, no. 2,pp. S56–S57, 2004. → page 29, 30[81] M. S. Gordon, “Cardiology patient simulator: development of an ani-mated manikin to teach cardiovascular disease,” The American journalof cardiology, vol. 34, no. 3, pp. 350–355, 1974. → page 30[82] D. Gillies and C. Williams, “An interactive graphic simulator for theteaching of fibrendoscopic techniques,” in Eurographics, vol. 1987, 1987,pp. 127–138. → page 30[83] B. Bray, C. R. Schwartz, D. L. Weeks, and S. Kardong-Edgren, “Hu-man patient simulation technology: perceptions from a multidisci-plinary sample of health care educators,” Clinical Simulation in Nurs-ing, vol. 5, no. 4, pp. e145–e150, 2009. → page 30126Bibliography[84] E. M. Benadom and N. L. Potter, “The use of simulation in train-ing graduate students to perform transnasal endoscopy,” Dysphagia,vol. 26, no. 4, pp. 352–360, 2011. → page 30[85] S. Lin, J. Chen, P. Hertz, and P. J. Kahrilas, “Dynamic reconstruc-tion of the orophanryngeal swallow using computer based animation,”Computerized medical imaging and graphics, vol. 20, no. 2, pp. 69–75,1996. → page 31[86] I. Scholten and A. Russell, “Learning about the dynamic swallowingprocess using an interactive multimedia program,” Dysphagia, vol. 15,no. 1, pp. 10–16, 2000. → page 31[87] “Oculus Rift via Wikimedia Commons,” Rift, accessed: 2015-07-14. → page 32[88] “A screenshot from the 3DiTeams game, player is percussing chestSource: Duke University Human Simulation and Patient Safety Centervia Wikimedia Commons,” percuss chest.JPG, accessed: 2015-07-14. → page 32[89] “Oculus Rift,”, accessed: 2015-07-14.→ page 33[90] “Sony’s Project Morpheus,”, accessed: 2015-07-14. →page 33[91] “Flexible endoscopy training with the Simbionix GI MentorTM,”, accessed: 2015-07-14.→ page 33[92] C. Williams, J. Baillie, D. Gillies, D. Borislow, and P. Cotton, “Teach-ing gastrointestinal endoscopy by computer simulation: a prototype forcolonoscopy and ercp,” Gastrointestinal endoscopy, vol. 36, no. 1, pp.49–54, 1990. → page 34[93] F. H. Harlow, J. E. Welch et al., “Numerical calculation of time-dependent viscous incompressible flow of fluid with free surface,”Physics of fluids, vol. 8, no. 12, p. 2182, 1965. → page 35127Bibliography[94] O. C. Zienkiewicz and R. L. Taylor, The finite element method: Solidmechanics. Butterworth-heinemann, 2000, vol. 2. → page 36[95] J. D. Anderson and J. Wendt, Computational fluid dynamics.Springer, 1995, vol. 206. → page 36, 66[96] H. U. Mair, “Review: hydrocodes for structural response to underwaterexplosions,” Shock and Vibration, vol. 6, no. 2, pp. 81–96, 1999.→ page37[97] D. J. Benson, “Computational methods in lagrangian and eulerian hy-drocodes,” Computer methods in Applied mechanics and Engineering,vol. 99, no. 2, pp. 235–394, 1992. → page 37, 38[98] C. Hirt, A. A. Amsden, and J. Cook, “An arbitrary lagrangian-euleriancomputing method for all flow speeds,” Journal of ComputationalPhysics, vol. 14, no. 3, pp. 227–253, 1974. → page 37, 38[99] J. O. Hallquist et al., “Ls-dyna theory manual,” Livermore softwareTechnology corporation, vol. 3, 2006. → page 38[100] G. Liu, K. Lam, and C. Lu, “Computational simulation of sympatheticexplosion in a high performance magazine,” in Proc. of 3rd WeaponEffects Seminar, Singapore, 1998, pp. 65–75. → page 38[101] R. A. Gingold and J. J. Monaghan, “Smoothed particle hydrodynamics:theory and application to non-spherical stars,” Monthly notices of theroyal astronomical society, vol. 181, no. 3, pp. 375–389, 1977. → page39, 110[102] L. B. Lucy, “A numerical approach to the testing of the fission hypoth-esis,” The astronomical journal, vol. 82, pp. 1013–1024, 1977. → page39, 109[103] J. Monaghan, “Smoothed particle hydrodynamics and its diverse ap-plications,” Annual Review of Fluid Mechanics, vol. 44, pp. 323–346,2012. → page 39[104] S. Li and W. K. Liu, “Meshfree and particle methods and their appli-cations,” Applied Mechanics Reviews, vol. 55, no. 1, pp. 1–34, 2002. →page 39128Bibliography[105] M. Sonomura, H. Mizunuma, T. Numamori, H. Michiwaki, andK. Nishinari, “Numerical simulation of the swallowing of liquid bo-lus,” Journal of Texture Studies, vol. 42, no. 3, pp. 203–211, 2011. →page 40[106] S. J. Cummins and M. Rudman, “An sph projection method,” Journalof computational physics, vol. 152, no. 2, pp. 584–607, 1999. → page40[107] J. Monaghan and R. Gingold, “Shock simulation by the particle methodsph,” Journal of computational physics, vol. 52, no. 2, pp. 374–389,1983. → page 40, 64[108] J. P. Morris, P. J. Fox, and Y. Zhu, “Modeling low reynolds numberincompressible flows using sph,” Journal of computational physics, vol.136, no. 1, pp. 214–226, 1997. → page 40, 65, 66, 67[109] G. Hopkin, “Neonatal and adult tongue dimensions,” The Angle or-thodontist, vol. 37, no. 2, pp. 132–133, 1967. → page 47, 48[110] J. E. Lloyd, I. Stavness, and S. Fels, “Artisynth: a fast interactivebiomechanical modeling toolkit combining multibody and finite ele-ment simulation,” in Soft tissue biomechanical modeling for computerassisted surgery. Springer, 2012, pp. 355–394. → page 49, 59, 70[111] “National dysphagia diet: Standardization for optimal care.” NationalDysphagia Task Force. American Dietetic Association., 2002. → page57, 72[112] I. Stavness, C. A. Sa´nchez, J. Lloyd, A. Ho, J. Wang, S. Fels,and D. Huang, “Unified skinning of rigid and deformable models foranatomical simulations,” in SIGGRAPH Asia 2014 Technical Briefs.ACM, 2014, p. 9. → page 60[113] L. Kavan, S. Collins, J. Zˇa´ra, and C. O’Sullivan, “Geometric skinningwith approximate dual quaternion blending,” ACM Transactions onGraphics (TOG), vol. 27, no. 4, p. 105, 2008. → page 61[114] J. VonNeumann and R. D. Richtmyer, “A method for the numerical cal-culation of hydrodynamic shocks,” Journal of applied physics, vol. 21,no. 3, pp. 232–237, 1950. → page 63129Bibliography[115] G.-R. Liu and M. B. Liu, Smoothed particle hydrodynamics: a meshfreeparticle method. World Scientific, 2003. → page 64, 68, 69, 101, 102,104, 106, 110, 116[116] J. J. Monaghan, “Simulating free surface flows with sph,” Journal ofcomputational physics, vol. 110, no. 2, pp. 399–406, 1994. → page 65,66, 67[117] M. Ferrand, D. Laurence, B. Rogers, D. Violeau, and C. Kassiotis,“Unified semi-analytical wall boundary conditions for inviscid, laminaror turbulent flows in the meshless sph method,” International Journalfor Numerical Methods in Fluids, vol. 71, no. 4, pp. 446–472, 2013. →page 70[118] A. K. Ho, L. Tsou, S. Green, and S. Fels, “A 3d swallowing simula-tion using smoothed particle hydrodynamics,” Computer Methods inBiomechanics and Biomedical Engineering: Imaging & Visualization,vol. 2, no. 4, pp. 237–244, 2014. → page 70[119] E. M. Bisch, J. A. Logemann, A. W. Rademaker, P. J. Kahrilas, andC. L. Lazarus, “Pharyngeal effects of bolus volume, viscosity, and tem-perature in patients with dysphagia resulting from neurologic impair-ment and in normal subjects,” Journal of Speech, Language, and Hear-ing Research, vol. 37, no. 5, pp. 1041–1049, 1994. → page 72[120] P. J. Kahrilas, S. Lin, J. Chen, and J. A. Logemann, “Oropharyngealaccommodation to swallow volume,” Gastroenterology, vol. 111, no. 2,pp. 297–306, 1996. → page 72[121] R. Hazelwood, M. Farazi, K. Humphries, S. Fels, and B. Martin-Harris,“Standardized mbss assessment clinician training patterns,” dysphagiaResearch Society (DRS), 2016 (Accpeted). → page 78[122] R. Likert, “A technique for the measurement of attitudes.” Archives ofpsychology, 1932. → page 85[123] G.-R. Liu, Meshfree methods: moving beyond the finite element method.Taylor & Francis, 2009. → page 106130Bibliography[124] J. J. Monaghan and J. C. Lattanzio, “A refined particle method for as-trophysical problems,” Astronomy and astrophysics, vol. 149, pp. 135–143, 1985. → page 110[125] J. P. Morris, “A study of the stability properties of sph,” arXiv preprintastro-ph/9503124, 1995. → page 110[126] G. R. Johnson, R. A. Stryk, and S. R. Beissel, “Sph for high velocityimpact computations,” Computer methods in applied mechanics andengineering, vol. 139, no. 1, pp. 347–373, 1996. → page 110131


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items