Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

In vivo measurement of absolute metabolite concentrations with quantitative magnetic resonance imaging… Graf, Carina 2019

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

Item Metadata

Download

Media
24-ubc_2019_september_graf_carina.pdf [ 2.44MB ]
Metadata
JSON: 24-1.0380533.json
JSON-LD: 24-1.0380533-ld.json
RDF/XML (Pretty): 24-1.0380533-rdf.xml
RDF/JSON: 24-1.0380533-rdf.json
Turtle: 24-1.0380533-turtle.txt
N-Triples: 24-1.0380533-rdf-ntriples.txt
Original Record: 24-1.0380533-source.json
Full Text
24-1.0380533-fulltext.txt
Citation
24-1.0380533.ris

Full Text

IN VIVO MEASUREMENT OF ABSOLUTE METABOLITE CONCENTRATIONS WITH QUANTITATIVE MAGNETIC RESONANCE IMAGING AND SPECTROSCOPY by  Carina Graf  B.Sc., Technische Universität Dortmund, 2016  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Physics)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  August 2019  © Carina Graf, 2019 ii The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the thesis entitled:  In vivo measurement of absolute metabolite concentrations with quantitative magnetic resonance imaging and spectroscopy  submitted by Carina Graf  in partial fulfilment of the requirements for the degree of Master of Science in Physics  Examining Committee: Dr. Cornelia Laule Supervisor  Dr. Erin L MacMillan Supervisory Committee Member  iii Abstract Magnetic resonance spectroscopy (MRS) measures relative signals arising from spins on different metabolites, e.g. N-Acetyl aspartate (NAA). To improve the interpretability of changes caused by disease, it is optimal to convert these relative signals to absolute concentrations e.g. by referencing it to the MR signal of water. Segmentation of high-resolution qualitative magnetic resonance images (MRI) is an accessible and easy-to-use method to estimate the properties of tissue water in the spectroscopic volume of interest (VOI), including water content, [H2O], and relaxation properties (T1, T2) with pre-determined literature values. However, these tissue properties can change in disease and with age. Therefore, we proposed the use of a quantitative MRI approach to reference metabolite concentrations by measuring subject-specific T1 and T2 relaxation as well as water content maps. The approach was first validated by measuring a range of biologically relevant water contents and metabolite concentrations in vitro. [H2O] was overestimated by 4.8% on average, while NAA concentrations were underestimated by 9.9%. In a study of ten healthy controls comparing the traditional segmentation quantification with the novel quantitative MRI method, we observed larger variabilities for subject-specific water properties, which did not propagate to the variability of the absolute metabolite concentrations of the neurochemicals (p > 0.37). Metabolite concentrations were lower with the quantitative MRI approach by -5.4% (p=0.002) in a white matter volume of interest (VOI) and -2.4% (p=0.002) in a grey matter VOI compared to the segmentation-based quantification. iv The quantitative MRI method for calculating absolute metabolite concentrations in MRS showed promising results, offering a potential alternative for the currently widely used segmentation approach.  v Lay Summary Magnetic resonance imaging (MRI) is a non-invasive technique to image the human body. It uses a strong magnet to generate images but can also be used to measure the concentration of biochemicals. In the past, the measured levels of chemicals were reported in relation to each other. This, however, can lead to misinterpretations of the underlying physiological mechanisms. Here we propose a novel method for reporting metabolite levels based on advanced MRI techniques and compare it to previously used reporting standards. We first measured the performance of the new method in samples of known metabolite levels and then compared it to a standard technique used when measuring chemicals in the human brain. The proposed method provided different metabolite levels when compared to a standard approach but showed a similar spread of values. The results show that group-level comparisons using the new technique are a viable alternative when reporting metabolite levels.  vi Preface This work was conducted at the UBC MRI Research Centre, BC, Canada. Together with Drs Cornelia Laule and Erin L. MacMillan, I designed and conceptualised the studies presented in Chapters 2 and 3. For Chapter 2, I planned and prepared Phantoms A and B after wet lab training by Dr Farah Samadi at the International Collaboration for Repair Discoveries (ICORD). I collected the data with the help of the technologists Laura Barlow and Neale Wiley, conceptualized and implemented an analysis pipeline, analysed the results and wrote the text with helpful guidance from Drs Cornelia Laule and Erin MacMillan. Part of the analysis code used in Chapters 2 and 3 was previously produced by Drs Sandra Meyers and Erin L MacMillan.  A version of Chapter 2 will be submitted for presentation at the 28th Annual Meeting of the International Society of Magnetic Resonance in Medicine (ISMRM). Graf, C., MacMillan, E.L., Wiley, N., Laule, C. Any in vivo data collected in Chapter 3 was covered by the University of British Columbia’s Clinical Research Ethics Board Certificate H03-70237, “Magnetic Resonance Measurements on Normal Volunteers”. I recruited and consented the participants (sometimes with the help of Poljanka Johnson) as well as scanned the volunteers under the supervision of a technologist or Dr Erin L. MacMillan. I performed all analyses and wrote the text, with input and guidance from Drs Cornelia Laule and Erin L. MacMillan. Portions of this chapter have been accepted and presented at the Annual Meeting of the International Society for Magnetic Resonance in Medicine (ISMRM): vii • C Graf, EL MacMillan, LR Barlow, IM Vavasour, C Laule, “Personalised Water Scaling: Quantifying absolute metabolite concentrations using measured tissue water content and attenuation maps”, 27th Annual Meeting of the International Society of Magnetic Resonance in Medicine, Poster Presentation, Montreal, QC, Canada, May 11-16, 2019. A version of Chapter 3 will be submitted for publication. Graf, C., MacMillan, E.L., Barlow L., Vavasour I.M., Laule, C. viii Table of Contents Abstract ............................................................................................................................... iii Lay Summary ......................................................................................................................... v Preface ..................................................................................................................................vi Table of Contents ................................................................................................................ viii List of Tables ....................................................................................................................... xiii List of Figures ...................................................................................................................... xiv List of Abbreviations ........................................................................................................... xvii Acknowledgements ........................................................................................................... xviii Dedication ........................................................................................................................... xix Chapter 1: Introduction ......................................................................................................... 1 1.1 The Human Central Nervous System .................................................................................. 1 1.2 Origin of the Nuclear Magnetic Resonance Signal ............................................................. 6 1.3 Relaxation ........................................................................................................................... 7 1.3.1 T1 ................................................................................................................................. 9 1.3.2 T2 ............................................................................................................................... 10 1.3.2.1 Biological Significance of T2 .............................................................................. 12 1.3.3 Basic Imaging Concepts............................................................................................. 12 1.3.3.1 Signal Generation through Spin Echoes............................................................ 12 1.3.3.2 Spatial Localisation through Magnetic Gradients ............................................. 13 1.3.4 Measurement of Relaxation Parameters .................................................................. 14 1.3.4.1 Measurement of T1 ........................................................................................... 14 ix 1.3.4.2 Measurement of T2 ........................................................................................... 15 1.3.5 Manipulation of Relaxation Parameters ................................................................... 17 1.4 Theory and Methods of MR Spectroscopy ....................................................................... 18 1.4.1 Origin of Characteristic Metabolite Spectra ............................................................. 18 1.4.1.1 Electron Shielding ............................................................................................. 19 1.4.1.2 Spin-Spin Coupling ............................................................................................ 19 1.4.2 Data Acquisition ........................................................................................................ 22 1.4.2.1 Pulse Sequences for MRS .................................................................................. 22 1.4.3 Measuring and Improving Spectral Quality .............................................................. 23 1.4.3.1 Signal-to-Noise Ratio (SNR) ............................................................................... 23 1.4.3.2 Chemical Shift Displacement Error (CSDE) ....................................................... 24 1.4.3.3 Linewidth (FWHM) ............................................................................................ 24 1.4.3.4 Water Suppression ............................................................................................ 25 1.4.3.5 Further Preparation Phases .............................................................................. 26 1.4.4 Data Processing ......................................................................................................... 26 1.4.4.1 Pre-processing................................................................................................... 27 1.4.4.1.1 Processing of Water Spectra ....................................................................... 27 1.4.4.1.2 Processing of Metabolite Spectra ............................................................... 27 1.4.4.2 Fitting ................................................................................................................ 28 1.4.4.3 Quantification of Metabolite Concentrations .................................................. 29 1.4.4.3.1 Referencing to External Standards ............................................................. 30 1.4.4.3.2 Referencing to Internal Standards .............................................................. 31 x Chapter 2: Phantom Measurements and Validations ............................................................ 37 2.1 Introduction ...................................................................................................................... 37 2.2 Methods ............................................................................................................................ 38 2.2.1 Phantom Preparation................................................................................................ 38 2.2.2 MR Experiments ........................................................................................................ 42 2.2.2.1 Phantom A: Relaxation Measurements ............................................................ 42 2.2.2.2 Phantom B: Validations ..................................................................................... 43 2.2.3 Image Analysis........................................................................................................... 45 2.2.3.1 Quantitative Relaxation and Water Content Mapping ..................................... 45 2.2.3.2 MR Spectroscopy Analysis ................................................................................ 46 2.2.3.3 Structural Analysis ............................................................................................ 47 2.2.3.4 Statistical Analysis ............................................................................................. 48 2.3 Results ............................................................................................................................... 48 2.3.1 Determining a Concentration for a T1 & T2 Shortening Agent.................................. 48 2.3.2 Water Content Mapping with a 48-echo GRASE Sequence...................................... 49 2.3.3 Accuracy of In Vivo Metabolite Concentrations ....................................................... 50 2.4 Discussion and Limitations ................................................................................................ 52 2.4.1 Sample Preparation .................................................................................................. 53 2.4.2 Phantom Design & Temperature .............................................................................. 54 2.4.3 MR Measurements ................................................................................................... 55 2.4.4 Accuracy of Water Content Measurements Using a 48 echo GRASE Sequence. ..... 56 2.4.5 Accuracy of NAA Concentrations .............................................................................. 56 xi 2.5 Conclusion ......................................................................................................................... 57 Chapter 3: Towards Personalised Water Scaling of In Vivo Metabolite Concentrations. ........ 59 3.1 Introduction ...................................................................................................................... 59 3.2 Methods ............................................................................................................................ 60 3.2.1 Sample Overview and Experimental Set Up ............................................................. 60 3.2.2 Main Study ................................................................................................................ 61 3.2.2.1 Participant Information .................................................................................... 61 3.2.2.2 Scan Protocol .................................................................................................... 61 3.2.3 CSF Sub-Study ........................................................................................................... 66 3.2.4 MRI and MRS Data Processing .................................................................................. 66 3.2.4.1 GRASE Analysis .................................................................................................. 66 3.2.4.2 IR-T1 Series ........................................................................................................ 67 3.2.4.3 Structural and Volumetric Analysis ................................................................... 67 3.2.4.4 SVS Analysis ....................................................................................................... 68 3.2.4.5 Calculation of Water Scaling Factors (T1, T2, RATTH2O and TWC) ........................ 69 3.2.4.6 Quality Assurance and Statistical Analysis ........................................................ 70 3.3 Results ............................................................................................................................... 71 3.3.1 CSF Flow Effect on T1 Relaxation .............................................................................. 71 3.3.2 TR Effects on T1 Mapping in CSF ............................................................................... 73 3.3.3 Correlation of Segmentation and T2 Based Estimates of Voxel CSF Volume Fraction ……………………………………………………………………………………………………………………………..74 3.3.4 Spectral Data Quality ................................................................................................ 75 xii 3.3.5 Relaxation and Water Scaling Factors ...................................................................... 78 3.3.5.1 Relaxation Correction ....................................................................................... 79 3.3.5.2 Water Content .................................................................................................. 82 3.3.6 Metabolite Concentrations ....................................................................................... 83 3.4 Discussion.......................................................................................................................... 85 3.4.1 Scan Time .................................................................................................................. 86 3.4.2 Reliability of T1-Mapping .......................................................................................... 86 3.4.3 Estimate of CSF Fractions .......................................................................................... 87 3.4.4 Water Content Measurements ................................................................................. 88 3.4.5 Absolute Metabolite Concentrations ....................................................................... 89 Chapter 4: Conclusion and Future Directions ........................................................................ 90  xiii List of Tables Table 2.1: Prepared concentrations of MnSO4/H2O solutions and corresponding mean relaxation values of test tubes with standard deviations (SD). .................................................. 39 Table 2.2: Prepared sample concentrations for in vitro phantom validations of Phantom B. ..... 41 Table 2.3: Imaging sequence parameters for measurements in Phantom B. .............................. 44 Table 2.4: MR Spectroscopy sequence parameters for four samples of Phantom B. .................. 45 Table 3.1: MR Imaging sequence parameters .............................................................................. 64 Table 3.2: MR Spectroscopy sequence parameters ..................................................................... 65 Table 3.3: Data quality and VOI composition consistency ........................................................... 76 Table 3.4: Mean correction constants for each volunteer in global white matter. ..................... 80 Table 3.5: Mean correction constants for each volunteer in cortical grey matter. ..................... 81 Table 3.6: Absolute concentrations and percent differences between the quantification methods in the central white matter (cWM) voxel. .................................................. 85 Table 3.7: Absolute concentrations and percent differences between the quantification methods in the parietal grey matter (pGM) voxel. .................................................... 85 xiv List of Figures Figure 1.1: Structural elements of the central nervous system. .................................................... 3 Figure 1.2: Chemical structures of highly concentrated metabolites in the human brain. ........... 5 Figure 1.3: Example slice for a T1-weighted (T1w) and (B) T2-weighted (T2w) brain MRI. .............. 9 Figure 1.4: Measurement of T1 by inversion recovery (IR). .......................................................... 15 Figure 1.5: Diagram of the CPMG sequence. ................................................................................ 16 Figure 1.6: (A) Chemical structure and (B) characteristic spectrum of alanine caused by electron shielding and J-coupling mechanisms. ....................................................................... 21 Figure 1.7: (A) PRESS sequence and (B) localisation scheme. ...................................................... 23 Figure 1.8: Example of metabolite basis set (A) and LCModel fit (B). .......................................... 28 Figure 1.9: Flowchart for quantifying metabolite concentrations with reference to tissue water. .................................................................................................................................... 34 Figure 2.1: Schematic of Phantom A. ............................................................................................ 40 Figure 2.2: Schematic of Phantom B. ............................................................................................ 42 xv Figure 2.3: Measured water relaxation constants of Phantom A for variable concentrations of manganese sulfate monohydrate in deionized water. .............................................. 49 Figure 2.4: Accuracy of water content, [H2O], measurements in vitro. ....................................... 50 Figure 2.5: Representative spectrum from Sample X for Phantom B. ......................................... 51 Figure 2.6: Accuracy of NAA concentrations, [NAA], measurements in vitro. ............................. 52 Figure 3.1: Scan-rescan reproducibility of T1 maps of CSF (left, middle) and impact of phase encoding directions on T1 measurements of CSF in single individual (left, right). .... 72 Figure 3.2: Histograms of ventricular T1 values. ........................................................................... 73 Figure 3.3: Histograms for four volunteers showing the effect of different shot intervals (blue = 6s, orange =3s) and inversion times on the measurement of T1 in CSF. ................... 74 Figure 3.4: Correlation of CSF contributions identified by segmentation (x-axis) or T2 time (y-axis) within the pGM VOI for each volunteer. ........................................................... 75 Figure 3.5: Pre-processed spectra for both VOIs averaged across ten healthy controls (black). 77 Figure 3.6: Sample T1 (A), T2 (B), RATTH2O (C), and [H2O] (D) maps for one subject. ...................... 78 Figure 3.7: Water relaxation attenuation for both quantification methods. ............................... 82 xvi Figure 3.8: Water content correction factors for both quantification methods. ......................... 83 Figure 3.9: Absolute metabolite concentrations as calculated with SEG (orange) and TWC (blue). .................................................................................................................................... 84   xvii List of Abbreviations ap Anterior - posterior CSF Cerebrospinal fluid cWM Central white matter fCSF Partial volume fraction of CSF fGM Partial volume fraction of grey matter fh Foot - head FOV Field of view fWM Partial volume fraction of white matter Glu Glutamate GM Grey matter GRASE Gradient and spin echo Ins myo-Inositol MnSO4.H2O Manganese sulfate monohydrate MP-RAGE Magnetisation prepared rapid gradient echo MRI Magnetic resonance imaging MRS Magnetic resonance spectroscopy NAA N-Acetyl aspartate NMR Nuclear magnetic resonance NSA Number of spectra acquired pGM parietal grey matter PRESS Point- resolved spectroscopy RATTH2O Water signal attenuation rl Right - left  SENSE Sensitivity encoding SVS Single voxel spectroscopy T1 Spin-lattice relaxation time T2 Spin-spin relaxation time tCho total Choline tCr total Creatine TE Echo time TI Inversion time TR Repetition time WC or [H2O] Water content WM White matter    xviii Acknowledgements My sincerest thank belongs to my supervisors, Dr Cornelia Laule and Dr Erin L MacMillan for their continued knowledge, guidance and unlimited support throughout the past years. I could always rely on their mentorship through successful and less successful times. Dr Irene Vavasour provided me with invaluable help and always had an answer for my questions for which I am grateful. I would also like to thank the team of the UBC MRI Research Centre, especially technologist supervisor Laura Barlow for providing me with the training to operate the MR Scanner and technologist Neale Wiley for assisting during phantom scan and providing helpful suggestions. I sincerely thank Drs Alex MacKay and David Li for their valuable feedback and questions during my graduate research. I thank my friends and colleagues, Poljanka Johnson and Stephen Ristow, for their wonderful research help and my good friends Lisa Lee and Sarah Morris for an open ear in challenging times. My friends, close and far, who I have been fortunate enough to meet and learn to rely on, have been a resource I would not want to miss. My sincerest gratitude to my wonderfully patient volunteers who donated their time to my research. Lastly, I would like to thank the MS Society of Canada for allowing me to pursue this degree at the University of British Columbia by awarding me an endMS Master’s Studentship. xix Dedication   I dedicate this thesis to my Parents and Sibling for their unconditional Love and Support  1 Chapter 1: Introduction Magnetic resonance imaging (MRI) has become an invaluable diagnostic tool in medicine, primarily due to the superb soft tissue contrast it provides and the absence of any ionising radiation. MRI has thus grown into an essential resource for imaging the human brain and spinal cord, the main constituents of the central nervous system (CNS). This thesis will commence with a brief overview of essential neuroanatomy of the CNS followed by a concise introduction of the most necessary physical principles and techniques of MRI and magnetic resonance spectroscopy (MRS) as relevant to this thesis. This information has previously been covered in any standard work on MRI and MRS (1–4). For a more in-depth discussion, the reader is referred to excellent texts written by Levitt (1), Brown et al. (2) and de Graaf (4).  1.1 The Human Central Nervous System The human CNS consists of the spinal cord and the brain. The cerebrum is often referred to as the main brain, but only represents a single macroscopic component of the whole brain. It consists of four lobes (frontal, temporal, parietal, and occipital) which are separated by sulci and form gyri, whose primary function is to increase the surface area of the cortex (Figure 1.1-A). The cortex, also called grey matter (GM), due to its darker appearance post mortem, is comprised primarily of neuronal cell bodies and dendrites. The white matter (WM) on the other hand houses the axons of the neurons, around which cells called oligodendrocytes wrap a sheath called myelin (Figure 1.1-B).  2 Myelin consists of multiple lipid bilayers, which wrap around the axon to form an insulator that increases the conductivity of the axon and accelerates the signal transmission of action potentials along it (Figure 1.1-C). Thin layers of water are enclosed between the myelin membranes.  On the macroscopic scale, water is a crucial component of cerebrospinal fluid (CSF). The entire CNS is embedded in CSF, which consists primarily (99%) of pure water (5). Its function is to provide nutritional support, flush the brain of toxins, and provide cushioning between CNS tissue and the skull (6). While CSF surrounds the CNS, it also has larger pockets deep inside the cerebrum, termed ventricles. The most substantial pockets are the left and right lateral ventricles. CSF within the ventricular system is not stationary but pulsates with the cardiac rhythm, which is both measurable by MRI and influences the appearance of images (7). 3  Figure 1.1: Structural elements of the central nervous system. (A) Schematic structure of the cerebral anatomy with the three main compartments: white matter (WM), grey matter (GM) and cerebrospinal fluid (CSF) in the lateral ventricles (blue). (B) Drawing of a nervous cell, neuron, with a myelinated axon. GM primarily consists of the cell body and dendrites while WM receives its light appearance from the myelin lipid sheath wrapped around the axon. (C) Cross-section of a myelinated axon. Water molecules are present not only in the intra and extra-cellular spaces but also trapped within the myelin lipid bilayers. Images adapted from (8, License CC BY 3.0) and (9, License CC BY-SA 3.0). Besides being sensitive to stationary and pulsating water molecules, an MR scanner can also detect other small molecules which are dissolved in cell water and are present in a sufficiently VentriclesWhite MatterGrey Matter( A ) Macroscopic NeuroanatomyMyelin SheathAxonCell Body Axon TerminalWhite MatterGrey Matter Grey Matter( B ) The Neuron( C ) Micrograph ofMyelinated AxonIntra-cellular SpaceExtra-cellular SpaceMyelin LipidBilayersDendrite4 high concentration on the order of approximately >1 mmol/L. In the brain, the metabolites (Figure 1.2) which are most commonly measured include 1. N-acetyl aspartate (NAA), frequently interpreted as a neuronal marker (10)  2. Glutamate (Glu), the major excitatory neurotransmitter (11) 3. Total creatine (tCr), contributing to energy metabolism (12) 4. Choline-containing compounds (tCho), often found in membranes and 5. myo-Inositol (Ins), a glial marker (13). 5  Figure 1.2: Chemical structures of highly concentrated metabolites in the human brain. Note the abundance of hydrogen atoms within the molecular structures.  (A) N-acetyl-aspartate (NAA) (B) Glutamate (Glu) (C) Creatine (Cr) (D) Choline (Cho) (E) myo-Inositol (Ins) 6 1.2 Origin of the Nuclear Magnetic Resonance Signal Sixty-two percent of all atoms in the human body are hydrogen (1H) atoms (14, pp. 8, 664). These atoms are primarily located in water (H2O), lipids and other organic molecules (e.g. Figure 1.2). Since the nucleus of 1H only consists of an individual proton the overall nuclear spin, j, is non-zero, causing a magnetic moment, m. Due to quantisation, the z-component of the spin property of the proton only has two states it can occupy: spin “up” | ↑ ⟩ or spin “down” | ↓ ⟩. These states are indistinguishable under normal circumstances but their energy changes if the proton is placed in an external magnetic field. This phenomenon is known as the Zeeman effect. Furthermore, the external magnetic field, B0, also makes the proton’s spin precess around the axis of the static magnetic field with an angular frequency proportional to the strength of the magnetic field, the Larmor frequency ω:  𝜔 = 𝛾𝐵0 (1.1) The proportionality constant γ is called the gyromagnetic ratio. It can also be expressed in terms of the g-factor, gn, which differs between protons and neutrons as well as nuclei:  𝛾 =𝜇𝑁ℏ⋅ 𝑔𝑛 (1.2) where μN and ℏ are the nuclear magneton and the reduced Planck’s constant, respectively.  To minimise the total entropy of a system composed of a pool of protons, and to reach the thermal equilibrium, the lower energy level must be populated more frequently. Thus, it is slightly more probable that the protons’ spins align themselves parallel to the magnetic field, as 7 opposed to anti-parallel. Quantitatively speaking, the surplus of the protons’ spins which are aligned parallel to the magnetic field can be described by;  𝑅 = 𝑒𝛾ℏ𝐵0𝑘𝐵𝑇 ≈ 1 +𝛾ℏ𝐵0𝑘𝐵𝑇= 1 +Δ𝐸𝑘𝐵𝑇 (1.3) where R is the ratio and ΔE the energy difference between the two energy states, kB is the Boltzmann constant, and T the temperature. Equation (1.3) emphasises the importance of strong static magnetic fields, since the population difference between the two energy states increases continuously with B0. All spins combined build a net magnetisation, M0, of magnitude  𝑀0 =𝑁𝛾2ℏ2𝐵04𝑘𝐵𝑇 (1.4) where N is the number of excess spins in the lower energy state. 1.3 Relaxation A milestone for nuclear magnetic resonance (NMR) was achieved in 1946 when Felix Bloch mathematically described the behaviour of the magnetisation in a magnetic field with a single (vector) differential equation (15):  𝑑?⃗⃗? 𝑑𝑡= 𝛾?⃗⃗? × ?⃗? −𝑀𝑥?̂? + 𝑀𝑦?̂?𝑇2−𝑀𝑧?̂? + 𝑀0?̂?𝑇1 (1.5) where x,̂ ŷ and ẑ are unit vectors, M⃗⃗⃗  is the time-dependent magnetisation vector, and T1 and T2 characterise the relaxation behaviour of the magnetisation. This differential system of equations can be described by two separate, time-dependent equations: one to describe the behaviour of the magnetisation component parallel to the applied magnetic field, termed the longitudinal magnetisation, Mz; and one to describe the behaviour of the magnetisation 8 perpendicular to the applied magnetic field, which commonly combines the Mx and My component into the transverse magnetisation component MT  𝑀𝑧(𝑡) = 𝑀0 + (𝑀𝑧(0) − 𝑀0) ⋅ 𝑒−𝑡 𝑇1⁄  (1.6)                        𝑀𝑇(𝑡) = 𝑀0 ⋅ 𝑒−𝑡 𝑇2⁄  (1.7) If a radiofrequency (RF) pulse is applied with a flip angle α around the x or y-axis, the equilibrium magnetisation is tipped away from the z-direction and can, for instance, be tipped into the transverse plane with α = 90°. An RF pulse of this flip angle is typically referred to as excitation pulse, whereas RF pulses with a flip angle of 180° may either be inversion pulses (if M0 is tipped from +z to -z) or refocusing pulses (if M⃗⃗⃗  was excited within the transverse plane) leading to the refocusing of the spin dephasing due to small local variations in B0. Equations (1.6) and (1.7) describe the behaviour of the magnetisation after an RF excitation.  While Mz recovers with a characteristic time constant of T1 back to the equilibrium strength M0, the MT component of the magnetisation decays exponentially towards 0 with a time constant of T2. T1 and T2 are sample and B0 dependent time parameters that are used in MRI to generate contrast in images, e.g. allowing the distinction between GM, WM and CSF in the brain (Figure 1.3). They can be exploited to highlight specific tissues while nulling the signal from others (16). 9  Figure 1.3: Example slice for a T1-weighted (T1w) and (B) T2-weighted (T2w) brain MRI.  (A) An image generated with a magnetisation-prepared rapid gradient echo (MP-RAGE) sequence (17). T1w-images provide high contrast between WM, (appears bright) and GM and CSF (appear dark). (B) T2w-images highlight CSF, while GM and WM appear darker.  T1 and T2 have different microscopic origins which can partially explain discrepancies between the simplified model of the Bloch equations in Equations (1.6) and (1.7) and measurements in reality. 1.3.1 T1 The longitudinal relaxation behaviour is characterised by T1 and causes the recovery of the longitudinal magnetisation. On a sub-atomic scale, it can be explained by the interactions between the nuclei’s spins and their atomic environment or lattice. T1 is therefore also commonly referred to as the spin-lattice relaxation constant. Any RF pulse disturbing the magnetisation from its equilibrium state will increase the entropy of the system. To return to equilibrium (level of minimum entropy), spins will exchange energy with the atomic lattice at a characteristic rate. The rate of energy exchange depends on the absolute difference between the current longitudinal magnetisation and its difference from equilibrium. ( A ) T1-weighted ( B ) T2-weighted10 1.3.2 T2 In general, any hydrogen nucleus contributes to the MR signal measured with MRI; however only some of these are detectable with MR systems and thus MR visible. To be MR visible, protons must have large enough T2 constants on the order of 5 ms or above for a typical imaging sequence. Shorter T2 times are physically possible and often occur in protons which are not part of the aqueous pool, e.g. solids and semi-solids such as lipids, proteins and other macromolecules. The decay of the transverse magnetisation component is related to spin-spin interactions and are influenced by the water protons’ environment. T2 can conceptually be explained in a classical way but requires a rigorous quantum mechanical derivation to prove it (1). At this point, we shall limit the explanation to a classical example, which is sufficient for most biologically relevant tissues. Imagine any molecule with hydrogen atoms that is freely floating in water. These molecules form the aqueous pool and shall be restricted in a fixed volume V. Molecules of interest, e.g. water, have the freedom to stretch, rotate, tumble around or bounce off each other in this given space, and the correlation time τC characterises the rate at which these processes occur.  Fast movements of individual molecules will lead to shorter τC. A fast mobile proton within the volume, therefore effectively experiences a constant magnetic field, since field fluctuations induced by neighbouring protons are only shortly within range, resulting in shorter spin-spin interaction times, due to the fast movement of the proton. 11 Hypothetically, if the volume of the aqueous pool were to be decreased, the free motion of the molecules would be restricted due to increased interactions between the molecules and boundaries. This causes increases in τc and relays to the protons in the aqueous pool due to new and more prolonged interactions between the aqueous and the non-aqueous protons from molecules with very short T2 which are located at the boundary. Any protons in the aqueous pool are now affected by a time-varying magnetic field caused by the precessing protons in the proximity. Differences in the effective experienced magnetic field between different protons will introduce a phase difference due to slight deviations from the original Larmor frequency. Over time, the differences in phase between different protons accumulate and lead to an overall loss in phase coherence, characterised by T2. The effect on MT can be visualised when remembering its composition. MT is a sum over all protons’ spins, each with its own small individual contribution to the measurable magnetisation MT. The introduction of phase differences between the individual spins can be thought of as each spin represented by an arrow pointing in a slightly different direction in the transverse plane, similar to what a fan would look like. After summation over all individual spins with phase differences, Δϕ, the net magnetisation decreases quicker for shorter T2 since Δ𝜙(𝑡) ∝ 1 𝑇2⁄ . Differences in the spins’ phase can also be introduced by inhomogeneities in the magnetic fields due to external factors, e.g. insufficient shimming (cf. 1.4.3) leading to a quicker decay of MT than if only spin-spin interactions were present. Effects contributing to this quicker decay can be summarised in the characteristic decay time T2’: 12  1𝑇2∗ =1𝑇2+1𝑇2′ (1.8) The constant of T2* characterises the free induction decay (FID) in the transverse plane. While the signal decay due to T2 is irreversible, the signal may be partially or fully recoverable by an RF pulse if it originates from temporally non-varying magnetic field inhomogeneities. 1.3.2.1 Biological Significance of T2 In Section 1.1, we previously discussed the microscopic structure of the CNS and highlighted that water is present in different microscopic compartments. Water trapped within the myelin-lipid bilayers (myelin water) specifically exhibits a characteristically short T2 time (18) in comparison to intra- and extracellular water (IE-water) and CSF due to its confined space. Analysing a series of multi-echo images (Section 1.3.4.2) with a non-negative least squares algorithm (19) allows the separation of the measured signal fraction originating from water trapped within bilayers based on the short T2. The signal fraction attributed to the short T2 component is termed myelin water fraction (MWF) and has been validated to correlate with the phospholipids of myelin (20).  1.3.3 Basic Imaging Concepts To measure the magnetisation and/or decay of such with an MR scanner, a combination of RF pulses and magnetic gradient fields are applied. 1.3.3.1 Signal Generation through Spin Echoes Measurement of the magnetisation in an MR scanner is only possible in the transverse plane spanned by the x and y-direction, due to current hardware limitations. Thus, any measurement 13 sequence has to contain an excitation RF pulse with a flip angle such that the equilibrium magnetisation is either partially or fully tipped into the transverse plane. This, in theory, is enough to measure the FID of the magnetisation. However, magnetic field inhomogeneities as mentioned in Section 1.3.2 lead to a quick signal decay and could thus prevent the detection of metabolites with short T2*, since a specific time - the detector dead time - is required to switch between the transmission of RF pulses and detection of the FID. A solution to this is to append a second pulse to the initial excitation at time t which may either partially or fully rephase the signal at time 2t previously lost due to the T2* decay. This phenomenon of refocusing after 2t is known as an echo and, if the echo time is short enough, is of almost the same amplitude as the original FID (decreased to 𝑒−𝑇𝐸𝑇2). It can be shown that the generation of an echo does not depend on the flip angle of either RF pulse. However, in order to maximise the signal achievable by a spin echo, a refocusing pulse of 180° should be applied which refocuses all magnetisation lost due to T2* relaxation. This pulse sequence of any two RF pulses is referred to as the original Hahn Echo (21). Overall, the highest signal can be achieved when the flip angle of the first RF pulse is chosen to be 90°, and the consecutive 180°-pulse is used for refocusing so that a full excitation is followed by a complete refocusing of the magnetisation. 1.3.3.2 Spatial Localisation through Magnetic Gradients While applying RF pulses by themselves will generate a measurable signal, it is preferable if the exact origin or location of the signal can be specified. 14 Spatial localisation in MRI can be achieved by using the relationship between resonance frequency and the effective magnetic field. If we recall the Larmor equation (1.1), we remember that the resonance frequency ω is proportional to the static magnetic field B0. Thus, if a second spatially linear varying magnetic field is overlaid with the original B0, spins at different locations will have different resonant frequencies, and only the ones which fall into the spectral region excitable by a given RF pulse of central frequency ω and bandwidth Δω will be excited if the RF pulse and gradient are played in sync. This concept is known as slice selection and can be used for three-dimensional localisation. For MR imaging, the magnetisation signal can additionally be phase and frequency encoded to allow for improved time efficiency. 1.3.4 Measurement of Relaxation Parameters 1.3.4.1 Measurement of T1 The gold standard for measuring T1 relaxation is an inversion recovery (IR) spin echo sequence (Figure 1.4) where a 180° inversion pulse precedes the 90° excitation and signal readout. By repeating the experiment with different inversion times (TIs), T1 can be extracted from a modified Equation (1.6):  𝑀𝑧(𝑡) = 𝑀0(1 − 𝑓 ⋅ 𝑒−𝑇𝐼 𝑇1⁄ ) (1.9) Where f is 2 for an exact 180° pulse. In cases where f = 1 the inversion pulse is decreased to a 90° pulse. This shortens the acquisition time but leads to a smaller and potentially worse dynamic signal range. To reduce scan time, other T1 mapping approaches, including variable flip angle (VFA), can be used (22). This method maps T1 from a series of spoiled gradient recalled 15 (SPGR) echo images and is referred to as DESPOT1 (driven equilibrium single pulse observation of T1) (23).  Figure 1.4: Measurement of T1 by inversion recovery (IR). (A) The IR pulse sequence is indicating the timing of crucial sequence parameters. A spin echo is formed at the echo time (TE) due to the rephasing of the spins caused by the second 180° pulse. The echo strength (maximum signal intensity) is dependent on the inversion time (TI), and repetition time (TR). TR is the time between two excitation pulses and is the limiting factor for TIs. T1 can be probed by repeating the experiment with varying TIs. (B) Signal recovery of M0 after full inversion for a short (dark orange), medium and long T1 (light orange). T1 can be characterised by measuring at least two data points along these curves.  1.3.4.2 Measurement of T2 The time dependence of MT following an RF pulse has previously been described in Equation (1.7). When measuring MT at different time points, t, one can change the detectable signal. Thus, a common technique to measure the T2 of a specimen is by measuring a time series with varying TEs. The gold standard in multi-TE techniques is the Carr-Purcell-Meiboom-Gill (CPMG) sequence (Figure 1.5) (24, 25). Here, the initial 90° excitation pulse is followed by a train of one or more 180° refocusing pulses. It differs from a standard spin echo only slightly by means of the phases and directions in which the refocusing pulses are applied, with the ultimate goal of RF90° 180°SignalTETR90°180°TIM0-M00.26 M0x-axis: TI 0.5 T_1T12 T1½ 1½ T12 T1T1 2T1( A ) Inversion Recovery (IR) Pulse Sequence ( B ) Inversion Recovery SignalSignal16 minimising the intensity of stimulated echoes. These can only occur in a sequence of three or more RF pulses and arise when the flip angles of the first and second RF pulses “store” part of the magnetisation in the z-direction. The successive pulses then tip the stored magnetisation back into the transverse plane, where they can form an echo after rephasing (26).  Figure 1.5: Diagram of the CPMG sequence. Shown are RF pulses (top) for excitation and refocusing of the magnetisation and gradients (GSlice, GPhase and Greadout, middle) for spatial localisation of the signal (bottom) at certain times after the excitation. RF transmission (green) and receive times are interleaved. Each FID decays at a rate of T2* (red), which can be reversed by a refocusing pulse to form echoes (orange). A frequency selective readout gradient (Greadout) is played during readout periods of echo formation times TE1 and TE2. Note the decrease in echo intensities, which is not refocused, representing the signal decay due to T2 relaxation. While the CPMG may be the gold standard in measuring T2, it is not easily usable in vivo due to the long acquisition times. To accelerate data collection and increase coverage of anatomical regions, one may utilise a combined gradient and spin-echo (GRASE) sequence which shortens RFGSliceGReadoutGPhaseSignalTimex-90° y-180° y-180°0 TE1 TE2T2T2*17 acquisition time, through the use of two gradient echoes alongside each spin echo (27). The gradient echoes obtained with GRASE sequences could also be used to decrease the specific absorption rate (SAR) while retaining the same number of echoes through the echo-planar imaging (EPI) acceleration factor of 3. The main drawback of gradient echoes is their high sensitivity to differences in T2* caused by spatial variations of magnetic susceptibility or B0 inhomogeneities (27). This sensitivity is actively used when measuring T2* with gradient echo (GE) sequences (28).  Other techniques to measure T2 include variable flip angle approaches such as mcDESPOT (multi-component driven equilibrium single pulse observation of T1/T2 (29)), or sequences with a T2 preparation phase, followed by any readout sequence (e.g. spiral readout) (30). 1.3.5 Manipulation of Relaxation Parameters Occasionally, it is of interest to change the relaxation parameters of the investigated object. This may happen, e.g. when aiming to replicate T1 and T2 present in biological tissue in phantoms, either for validation or quality assurance purposes. In these cases, paramagnetic ions, e.g. Gadolinium (Gd3+), Copper (Cu2+), Manganese (Mg2+) or other contrast agents (CA) can be added to a solution to shorten both T1 and T2 through interactions between CA and water molecules (31).  The efficiency of a contrast agent is given by its relaxivity, ri. It quantifies the relationship of changes in the relaxation parameters R1 and R2 (the inverse of T1 and T2) and the molecular concentration of the contrast agent, [CA] (32): 18 𝑅𝑖 = 𝑅𝑖,0 + [𝐶𝐴] ⋅ 𝑟𝑖 (1.10) The main determining factors for relaxivities are the molecular structure of the CA and the molecule it interacts with as well as B0 and temperature (33).   1.4 Theory and Methods of MR Spectroscopy MRS has been around as long as NMR itself and was the original technique used to measure the free induction decay of the excited magnetisation. In vivo, it is of physiological interest since it allows the measurement of concentrations of metabolites that are small enough to be freely floating and rotating in water. This permits hydrogen nuclei within the molecules to have T1 and T2 relaxation parameters on the order of several milliseconds and are thus long enough to be measured with clinical and research MR scanners. Furthermore, as previously mentioned in Section 1.1, only metabolites which are present in a sufficiently high concentration (> 1 mmol/L) will be detectable. 1.4.1 Origin of Characteristic Metabolite Spectra Different molecules are distinguishable by their characteristic spectra in the frequency domain. There are two main processes that give rise to the hydrogen nuclei spectra. Firstly, there is electron shielding, which leads to a small shift in the resonance frequency of the measured proton, also referred to as chemical shift. Secondly, spin-spin or J-coupling causes the splitting of a resonance peak into a subset of smaller peaks. To explain these two processes, we must further investigate the sub-atomic and molecular structure and interaction of the chemicals. 19 1.4.1.1 Electron Shielding Consider a proton in the hydrogen nucleus. The effective static magnetic field it experiences not only depends on the external magnetic field B0 generated by the superconducting magnet but is also affected by the electron cloud surrounding the proton. In the Bohr model, the electron circulates the nucleus, thus creating an electric current, which induces an additional magnetic field, Be-. Depending on the direction of circular motion, Be- will either increase or decrease the magnetic field experienced by the nucleus, Beff. Furthermore, this magnetic field may vary depending on the electronegativity of the covalent bond in the molecular structure. E.g. the electron in an O-H group will be more attracted by the oxygen atom, leading to a weaker electron shielding of the hydrogen nucleus in comparison to a covalent C-H bond. Overall, different Beff s correspond to slight changes in the resonance frequencies on the order of parts per million (ppm). This chemical shift, δ, is typically calculated with respect to a reference chemical and expressed in ppm;  𝛿 (ppm) =𝜔 − 𝜔𝑟𝑒𝑓𝛾𝐵0⋅ 106 (1.11) 1.4.1.2 Spin-Spin Coupling The other interaction which contributes to the characteristic spectrum of each metabolite is spin-spin or J-coupling. It is based on the quantum mechanical phenomenon of spin interactions within a molecule and can act over a distance of up to three molecular bond lengths. It is best explained by considering a simple molecule, e.g. alanine (Figure 1.6-A). The three chemically and magnetically equivalent protons of the 3C-H3 group have a chemical shift of approximately 1.4 ppm (Figure 1.6-B). Due to J-coupling, the nuclear spins interact with 20 the electric spins and align each other in antiparallel pairs. In addition, the proton of the 2C-H group can be aligned either parallel or anti-parallel to the external magnetic field as previously discussed. Depending on its spin orientation, it changes the resonance frequency of the 3C-H3 protons and splits the single peak into a doublet peak with the area underneath the combined peaks equalling the area of the singlet. Simultaneously, each proton of the 3C-H3 group can be either parallel or anti-parallel aligned in comparison to the 2C-H proton. In total there are 23 possible combinations which can be summarised in four scenarios differing in the resulting resonant frequencies. 1. |↓↓↓⟩ 2. |↑↓↓⟩ 3. |↑↑↓⟩ 4. |↑↑↑⟩ While the former and latter options only have a single possible combination each, the middle scenarios can be generated through three separate constellations. Therefore, the 2C-H singlet resonance at 3.7 ppm will split into four distinguishable resonances with an intensity ratio of 1:3:3:1.  21  Figure 1.6: (A) Chemical structure and (B) characteristic spectrum of alanine caused by electron shielding and J-coupling mechanisms. The three chemically and magnetically equivalent protons of the methyl group (blue) give rise to a strong resonance at 1.4 ppm due to electron shielding and are split into a doublet by the interaction of J-coupling with the 2C-H proton (green). Vice versa, the 2C-H proton gives rise to the resonance at 3.7 ppm. Due to J-coupling with each of the methyl group’s protons, the signal of the 2C-H proton splits into four individual resonances. By convention, the x-axis of a spectrum is shown in reverse with larger frequencies on the left. Together, chemical shifts and scalar coupling generate unique spectra for each metabolite allowing the distinction between metabolites in a spectrum with overlapping resonances (34). In vivo spectra of proton MRS are highly overlapping since many molecules have similar structures e.g. glutamine (Gln) and glutamate (Glu) or N-acetyl-aspartate (NAA) and N-acetyl aspartyl glutamate (NAAG) but very different physiological functions. At 3T a reliable quantification of at least five metabolites can be achieved in the majority of single voxel spectra when acquired at short TEs. These are NAA, Glu, tCr, tCho and Ins (Figure 1.2). 123f( A ) ( B )22 Specialised pulse sequences, including J-editing, can target a larger variety of metabolites which are more challenging to detect due to lower concentrations or highly coupled spectral patterns e.g. for the detection of γ-Aminobutyric acid (GABA), the major inhibitory neurotransmitter (35). 1.4.2 Data Acquisition 1.4.2.1 Pulse Sequences for MRS Localisation sequences for MRS use slice selective duos of RF pulse and gradients. Each duo of RF pulse and gradient will excite an individual slice, thus if gradients are applied in each of the three perpendicular directions (x, y and z), a single RF pulse and gradient duo will excite a slice, the second perpendicular RF and gradient duo will only excite a column and lastly a third RF pulse and gradient duo will limit the excited and refocused signal to a volume of interest (VOI) origination from a cuboid. For a flip angle sequence of RF 90°-180°-180°, this sequence is known as Point RESolved Spectroscopy (PRESS, (36)) and permits the full refocusing of the magnetisation allowing high SNR (see Figure 1.7).  To minimise RF imperfections, phase cycling is typically used in the acquisition, requiring either 8 or 16 individual spectra to be acquired to average out phase differences introduced by said RF pulses. 23  Figure 1.7: (A) PRESS sequence and (B) localisation scheme.  (A) A 90° RF pulse followed by two 180° RF pulses allows for full refocusing of the magnetisation at the echo time (TE). The repetition time (TR) is the time between two excitation pulses. The three gradients (Gx, Gy, Gz) each select a slice in three orthogonal planes which result in a selected volume of interest (VOI) in the shape of a cuboid (B). Image sources: (B) adapted with permission from Erin L MacMillan. 1.4.3 Measuring and Improving Spectral Quality 1.4.3.1 Signal-to-Noise Ratio (SNR) Signal-to-noise in MRS is limited due to the low concentration of metabolites that are small enough, freely floating in water, and thus MR visible. Thus, one typically requires more than one acquisition to achieve a sufficiently high SNR. Acquisitions are repeated after a fixed repetition time, TR, which is ideally chosen to be greater than four times the longest T1 of the sample investigated (T1,max), to allow for a full recovery to the equilibrium state between consecutive excitations. This minimises T1-weighting of the metabolites’ signal amplitudes. The number of spectra acquired, NSA, affects the SNR with a factor of √NSA. Thus, while averaging RFx-90° y-180° y-180°SignalGzGyGxTETRx-90°( B ) Localization Scheme( A ) PRESS Sequence24 more acquisitions is beneficial, its SNR per time efficiency is limited. An alternative to increasing SNR is an increased voxel size since SNR is also directly proportional to each of the dimensions Δx, Δy and Δz of the voxel. 1.4.3.2 Chemical Shift Displacement Error (CSDE) Chemical shift displacement refers to the phenomenon of signals with different resonating frequencies, and consequently, different chemical shifts, originating from different regions in space. It is inversely proportional to the bandwidth of the RF pulses. Small CSDEs are especially desirable in in vivo measurements of voxels with heterogeneous tissues. Thus, pulse sequences with smaller flip and/or refocusing angles, e.g. a STimulated Echo Acquisition Mode (STEAM, (37)) sequence with an RF scheme of 90°-90°-90°, have smaller CSDE but come at the cost of only refocusing half of the original signal. CSDE can also be decreased through the application of adiabatic pulses with broad bandwidths as found in sequences such as Image-Selected In vivo Spectroscopy (ISIS, (38)), localisation by adiabatic selective refocusing (LASER, (39)) or semi-LASER acquisition schemes (40), which require high maximum RF amplitudes. 1.4.3.3 Linewidth (FWHM) Achieving high spectral quality in MRS also depends on the spectral linewidth. In order to confidently separate overlapping peaks at clinically available field strengths of 1.5 or 3T, the linewidth, which is characterized by the full width at half maximum (FWHM) of a peak, should be as narrow as possible. While its lower limit is determined by any metabolites’ T2, the linewidth is usually broader than that caused by pure T2 due to B0 inhomogeneities. Thus, for narrower linewidths, the magnetic field in the volume of interest must be homogenized. This 25 process is called active shimming and uses special shim coils producing small magnetic fields based on spherical harmonics to compensate for inhomogeneities in the static magnetic field. Active shimming must be repeated for every volume of interest in each participant since any inhomogeneities strongly depend on the participant’s anatomy. 1.4.3.4 Water Suppression While analogue-to-digital converters used in modern MR systems have the dynamic range to accurately digitise the broad dynamic range of a combined water and metabolite signal, it is often preferable to suppress the large water signal, which is up to 10 000 times stronger than that of the metabolites. Furthermore, baseline distortions and transient signals from mechanical vibrations of the hardware inside the MRI scanner are also reduced when suppressing the central water peak. Water suppression may be achieved by various pulse sequences implemented either before or interleaved with, the spatial excitation and localisation. A popular option for the suppression of water is a frequency selective excitation of the water signal followed by a magnetic gradient crusher which dephases the magnetisation in the transverse plane. To increase the efficiency of this method, the basic building block (RF pulse – crusher) is repeated several times (usually three iterations). The efficiency can be optimised by adjusting the flip angle of the third RF pulse such that the longitudinal magnetisation of the water signal is minimal at the time of the RF excitation pulse. 26 A variant of this method is the vendor-specific “Excitation” option available on Philips systems (Best, The Netherlands), for which the magnetisation is tipped slightly beyond the transverse plane followed by a crusher gradient dephasing the transverse magnetisation. The remaining longitudinal magnetisation recovers to Mz = 0 during a short delay time at which the acquisition of the spectrum begins.  1.4.3.5 Further Preparation Phases For the pre-scan routine to be successful, not only parameters for the shim and water suppression are optimised but also further factors which aid in detecting a signal. These include steps for improving the RF transmission and reception. To achieve maximum signal and minimise unwanted signal contributions due to imperfect RF pulses, the optimal power or amplitude for the RF pulses must be determined. This process is called power optimisation and ensures that the determined flip angles can be accurately achieved. Furthermore, since the water signal and metabolites’ signal are different on several orders of magnitude, the receiver gain must be adjusted, changing the dynamic range of the analogue to digital converter. 1.4.4 Data Processing While most scanner platforms have basic functionalities in analysing the spectra acquired, post-acquisition offline processing is preferable to maximise information extractable from each spectrum. Standardised data processing routines also allow for a comparison between subjects and different MRI sites and can aid in data interpretation. Data processing can be summarised into three main phases:  27 1. Pre-processing 2. Fitting 3. Quantification; each of which has several sub-processes. 1.4.4.1 Pre-processing Pre-processing of MRS data can be a valuable step to obtain or maintain good data quality. To perform the subsequent steps, it is desirable that non-water suppressed spectra are acquired, and each individual acquisition (each FID) of metabolite and non-water suppressed spectra are saved separately. The two acquisitions only differ in the presence (metabolites spectra) or absence (water spectra) of the RF pulses used for the water suppression 1.4.4.1.1 Processing of Water Spectra The non-water suppressed spectra must be acquired with the identical sequence as used for the consecutive metabolite spectra. The signal acquired from water spectra can be used to characterise spectral distortions, e.g. sidebands or phase distortions, due to the larger signal. Corrections to the metabolite spectra using the water reference spectrum include a zero and first-order phase correction as well as eddy currents, induced by gradients since these are expected to be identical between metabolite and water spectral acquisitions (41). 1.4.4.1.2 Processing of Metabolite Spectra The quality of the final spectrum used for fitting and quantification may also greatly be improved if individual acquisitions or FIDs are saved for pre-processing. It allows the alignment 28 of individual acquisitions in the frequency or time domain, thus minimising signal loss due to incoherent averaging and improving the linewidth of the resulting spectrum. Small differences in frequency can be considered normal due to physiological motion and do not warrant exclusion from the data set. 1.4.4.2 Fitting After pre-processing, spectra may be fitted with a fitting algorithm. Fitting of in vivo proton spectra typically utilises so-called basis sets for the metabolites which are expected to be present in the volume of interest to account for the highly overlapping spectra of the metabolite signals in short-TE MRS. A metabolite basis set and an example brain spectrum is shown in Figure 1.8.  Figure 1.8: Example of metabolite basis set (A) and LCModel fit (B). (A) Example of a metabolite basis set for brain MRS and its contributions to the spectrum. (B) Example of an LCModel fit for a brain metabolite spectrum from grey matter (collected with PRESS TE/TR=30 ms/4000 ms) with an artefact (orange). The residual resonance provides evidence of ethanol contamination in the subject. 0.20.71.21.72.22.73.23.74.2Chemical Shift (ppm)0.21.22.23.24.2Chemical Shift (ppm)( A ) ( B )29 The most frequently used fitting process minimises the residuals between the sum of linearly scaled metabolite intensities and the acquired spectrum as used, e.g. in LCModel (42). While LCModel measures residuals in the frequency domain, other algorithms, e.g. TARQUIN (43) fit the spectrum and minimise the residuals in the time-domain. Spectra which only have few prominent non-overlapping peaks, for instance when acquired at long TEs, may be quantified without the use of a basis set, by determining the signal intensity of each individual resonance.  The fitted spectra should always be visually inspected for artefacts, which would not be detectable in quantitative measures of spectral quality (e.g. SNR and FWHM). In the example given in Figure 1.8-B the spectrum provided evidence of a metabolite present which had not been included in the basis set and thus led to non-uniform residuals around 1.2ppm. The metabolite was identified to be ethanol. 1.4.4.3 Quantification of Metabolite Concentrations MRS Metabolite concentrations can be reported in different ways. Since the measured signal is in arbitrary units, the signal of an individual metabolite generally needs to be referenced to a second signal. The real physical signal intensity of this reference signal varies and can be determined either by using external standards such as same-place phantoms, electronic referencing, or using the drive scale factor (44). If these options are not feasible, referencing to internal standards, e.g. tissue water or other metabolites in the spectrum may be appropriate and yield similar results (45).  30 1.4.4.3.1 Referencing to External Standards When time permits, using the signal from an external reference may be the method of choice for metabolite concentration measurement. For one option involving external phantoms, one placed a phantom with a known concentration of a given metabolite, e.g. a large water bladder, close to the subject and conducted an acquisition localized within the phantom with the same parameters as the in vivo acquisition (same-time phantom) (46). Difficulties for this technique may arise from B1 inhomogeneities (both transmit and receive), the differences in internal temperatures of the samples, and the confined space inside typical coils, which may make the subject feel less comfortable. An improvement to this approach is the same-place phantom replacement method. It requires an identical MRS acquisition in the external phantom after the subject has been removed from the scanner (47). While it compensates for spatially varying B1 inhomogeneities if an appropriate phantom is chosen, it requires substantial extra time for acquisition, and results may be confounded due to differences in coil loading (48). It furthermore constrains the order of scans within the MR protocol, since in vivo and ex vivo scans must be collected consecutively and without repeating preparation phases. It has also been proposed to utilise an artificially injected reference signal into the receiver coil with a resonant frequency far away from the spectral information (49). This method can compensate for differences in coil loading by being calibrated with an external reference standard once. This approach, named Electronic REference To access In vivo Concentrations (ERETIC), however, does not intrinsically compensate for B1 differences and furthermore 31 requires additional hardware which can revoke a health authority’s approval due to custom coil modifications. Based on similar principles as the same-place phantom, a quantification technique termed the Drive Scale (DS) method was developed more recently (44). It addresses limitations, e.g. differences in coil loading, by compensating for variations in receiver gain, power optimization, as well as differences in both transmit and receive B1. The in vivo signal is corrected with a B1+ map and a contrast minimized image. It also requires the documentation of the power optimisation. The in vivo signal is then referenced to a phantom calibration measurement which only needs to be repeated occasionally as part of the quality assurance protocol, but does not require an extra acquisition with identical preparation phases in comparison to the traditional same-place phantom. 1.4.4.3.2 Referencing to Internal Standards In general, the most commonly used techniques are based on reporting the ratio of two metabolites (50). Historically, the most popular referencing metabolite was creatine (tCr) since it was conveniently assumed not to change in disease (51, 52), and the ratios of [met]/[tCr] are easily obtainable from any MRS fitting routine. Its primary benefit is that it does not require any additional scan time since it only uses information from the metabolite spectrum itself. Since it combines two individual measurements, it is also convenient in cases where the direction of change for each metabolite is known, e.g. NAA will decrease in MS, while tCho increases. Thus a ratio of the two metabolites will be more sensitive to changes and thus require a smaller sample size, which is advantageous for clinical trials and other economic purposes (53). While 32 this technique is undeniably convenient, more recent studies have shown that creatine can and does, in fact, change in pathology (12, 54) and varies across the brain (55). Therefore, if used as a reference, creatine ratios may mask the actual direction of changes in metabolites. If additional, non-water suppressed spectra were acquired, several further options for quantifying and reporting metabolite concentrations become available. Fitting routines such as LCModel allow the use of these additional spectra for referencing metabolite concentrations to the separately acquired water signal. Thus, instead of reporting [NAA]/[tCr] one would report [NAA]/[H2O]. If this format of reporting is chosen, it is typically referred to as institutional units (56). While it is an improvement in comparison to creatine ratios, institutional units are difficult to use if comparisons between subjects or across time are of interest. Furthermore, most VOIs will have different tissue compositions with varying fractions of white and grey matter as well as CSF, which causes variations in the referenced water concentrations and thus confounds the reported results. To account for this limitation, it has been proposed to estimate the tissue water content based on mean water concentrations for WM, GM and CSF and the voxel tissue composition obtained from high-resolution structural images (e.g. 3DT1-MPRAGE). A schematic of how tissue water concentration and attenuation can be estimated is shown in Figure 1.9-A. The structural image is first automatically segmented into WM, GM and CSF before the spectroscopic VOI is mapped onto the segmentation to extract the fractional VOI composition of each tissue type, fi. Literature values for tissue specific water content, [H2O]i, are then used 33 to estimate the water content in the VOI, [H2O]SEG, by weighting water content with corresponding tissue fraction: [𝐻2𝑂]𝑆𝐸𝐺 = 𝑓𝑊𝑀 ⋅ [𝐻2𝑂]𝑊𝑀 + 𝑓𝐺𝑀 ⋅ [𝐻2𝑂]𝐺𝑀 + 𝑓𝐶𝑆𝐹 ⋅ [𝐻2𝑂]𝐶𝑆𝐹 = ∑ 𝑓𝑖 ⋅ [𝐻2𝑂]𝑖                               𝑖∈𝑊𝑀,𝐺𝑀,𝐶𝑆𝐹     (1.12) The used literature values should come from a population similar to the sample investigated.  Tissue fractions can also be used to correct for relaxation attenuation of the water signal, RATTH2O. RATTH2O in spectroscopy is non-negligible due to reasonably long echo times of TE ≥ 20 ms and often the use of short TRs of less than 4s. Correcting for RATTH2O requires knowledge of T1 and T2 water relaxation times and can be explicitly calculated to be   𝑅𝐴𝑇𝑇𝐻2𝑂 =𝑆𝐻2𝑂(𝑀𝑅𝑆)𝑆𝐻2𝑂,0 = 𝑒−𝑇𝐸𝑇2 ⋅ (1 − 𝑒−𝑇𝑅𝑇1 ) (1.13) For a heterogenous VOI with known tissue fractions and tissue-specific T1 and T2 water relaxation, RATTH2O is defined as 𝑅𝐴𝑇𝑇𝐻2𝑂,𝑆𝐸𝐺 = ∑ 𝑓𝑖 ⋅ 𝑒−𝑇𝐸𝑇2,𝑖 ⋅ (1 − 𝑒−𝑇𝑅𝑇1,𝑖)𝑖∈𝑊𝑀,𝐺𝑀,𝐶𝑆𝐹 (1.14) Alternatively, a subject-specific T1 and T2 value can be measured with the acquisition of extra spectra with variable TEs and TRs. Ultimately, absolute metabolite concentrations, [met], can be calculated, giving the concentration in units of mmol/L (57) 34 [met] (mmolL) = [met]𝑓𝑖𝑡𝑡𝑖𝑛𝑔 ⋅ [𝐻2𝑂] ⋅ 𝑅𝐴𝑇𝑇𝐻2𝑂 (1.15) Where [met]fitting is the concentration ratio of a metabolite signal in reference to the water signal, determined, e.g. with LCModel.  Figure 1.9: Flowchart for quantifying metabolite concentrations with reference to tissue water. (A) Quantification by segmentation of high-resolution anatomical images (3D-T1 MPRAGE). Voxel tissue fractions, fi, together with tissue specific literature values provide an estimate for the water scaling factors of water content [H2O]SEG and RATTH2O,SEG required for reporting absolute metabolite concentrations in units of mmol/L. (B) Quantification with the proposed total water content method using quantitative MRI acquisitions. A combined GRadient And Spin Echo (GRASE) sequence maps water T2 and allows for signal extrapolation to TE=0ms, resulting in an S0 map. An inversion-recovery (IR) series maps water T1, which is required for the calculation of the total water content (TWC) and water relaxation attenuation, RATTH2O, map. The localisation of the spectroscopic VOI then permits the calculation of the mean water content, [H2O]TWC, and water relaxation, RATTH2O,TWC required for scaling the metabolite concentration [met]TWC according to (1.12) expressed in mmol/L. ( A ) SEGmentation Quantification ( B ) Total Water ContentQuantificationTWC map 5RATTH2Omap 𝐻2𝑂 𝑇𝑊𝐶[met]TWC(mmol/L)T1 mapT2 map𝑅𝐴𝑇𝑇𝐻2𝑂,𝑇𝑊𝐶3D multi-echo GRASE3D IR-T1 SeriesS0 map3D-T1MPRAGEBinary Voxel MaskSegmented ImagefCSF, fWMand fGM[met]SEG(mmol/L)Short-TE PRESS 𝐻 𝑂 𝑆𝐸𝐺  𝑅𝐴𝑇𝑇𝐻2𝑂,𝑆𝐸𝐺Fitting Results(6)Literature values for T1, T2 & [H2O] of CSF, WM and GM35 This method can be successfully used with only minimal additional scan time (approximately 6 mins for imaging) and allows for both longitudinal and inter-subject comparison in cohorts where global water content and tissue relaxation of water are not expected to change, and population averages in the different tissue types are pre-determined. Example cohorts include multiple sclerosis, as well as depression, schizophrenia and some brain tumours (58–60). Tissue water content and relaxation properties of water, however, may change in disease or healthy ageing, negating the assumption of constant tissue-specific water concentrations for the entire cohort (61). This warrants an explicit measurement of total water content (TWC) and water relaxation for each participant. Previous spectroscopy studies have used a T2* gradient echo-based or T1 relaxation-based water content mapping technique, which allowed the measurement of TWC within 7 mins (62, 63). While these techniques compensate for variable water content, they do not permit the correction of the relaxation attenuation of the water signal (RATTH2O) due to the water signal’s relaxation properties. More recently Meyers et. al (64) developed a T2 relaxation-based approach for measuring total brain water content (Figure 1.9-B), by extrapolating the T2 decay curve of a multi-echo GRASE sequence to TE=0 ms, producing a quantitative map for which signal intensity, S0, is directly proportional to water content (65). Part of the acquisition scheme is also a T1-mapping sequence, required for the calibration. This MR protocol thus provides information about water content as well as T1 and T2 of water and could be completed within 20 mins (66).  36 Since both T1 and T2 furthermore offer valuable physiological information, my thesis investigates the feasibility of measuring subject specific water scaling factors, [H2O] and RATTH2O, with a T2 relaxation-based approach to calculate absolute metabolite concentrations for MRS using the approach previously described by Meyers et al. (64). 37 Chapter 2: Phantom Measurements and Validations 2.1 Introduction The T2-relaxation based approach for water content mapping has previously been validated in simulations and phantom measurements (65), using both the traditional Spin-Echo and a faster 32-echo gradient and spin echo (GRASE) sequence. The diverse use of multi-echo relaxation data, e.g. for the mapping of myelin water fraction, has promoted the development of an updated GRASE sequence. With the shortest TE and an equal TE spacing of 8 ms, this 48-echo GRASE sequence offers an accelerated method to measure the characteristically short T2 component of myelin water with a T2 of approximately 20 ms (18). This novel sequence benefits from a reduced scan time (7.5 mins in comparison to previously 14.4 mins) while maintaining the same reconstructed resolution (67). However, the impact of shorter echo spacing and increased number of echoes in the new GRASE sequence on quantitative water content mapping is unknown and thus requires repetition of experiments similar to those previously conducted by S. Meyers (65).  The specific aims of this chapter are to (1) determine the accuracy of water content and relaxation mapping with the new 48-echo GRASE sequence and (2) validate the application of individual water content and relaxation mapping methods to calculate absolute metabolite concentration for MRS, by comparing measured and corrected metabolite concentrations to a known ground truth using in vitro phantoms.  38 2.2 Methods 2.2.1 Phantom Preparation Two phantoms were prepared to investigate the accuracy of the proposed water scaling method for referencing metabolite concentrations. Phantom A consisted of 11 sterile, polypropylene centrifuge tubes (16mL, Corning Inc., NY 14831) which were filled with various concentrations of Manganese Sulfate Monohydrate (MnSO4.H2O, Fisher Scientific) dissolved in deionised water (d-H2O) (68). To improve the accuracy of the final concentrations, a highly concentrated MnSO4.H2O solution of 0.112 mg/mL was diluted with pure d-H2O in various ratios with target concentrations ranging from 0.00 mg/ml to 0.06 mg/ml. The final concentrations of the 11 test tubes can be found in Table 2.1. All test tubes were sealed and submerged in a container filled with water to satisfy the minimum required coil loading for RF amplification. The resulting homogenous liquid water environment furthermore benefits B0 homogenization. A schematic drawing of Phantom A is shown in Figure 2.1. 39 Table 2.1: Prepared concentrations of MnSO4/H2O solutions and corresponding mean relaxation values of test tubes with standard deviations (SD). Test tubes 4 through 7 represent relaxation constants present in healthy human brain tissue. The segmentation method introduced in section 1.4.4.3.2. scales metabolite water ratios with water relaxation constants which are most closely resembled by test tube 6. Therefore, tube 6 was the concentration of choice for Phantom B. Test tube # [MnSO4.H2O] (mg/ml) T1 (s) (mean ± SD) T2 (ms) (mean ± SD) 1 0.000 3.25 ± 0.27 2381.7 ±  557.3 2 0.003 2.25 ± 0.08 400.3 ±    31.4 3 0.008 1.61 ± 0.09 183.3 ±    18.4 4 0.013 1.23 ± 0.05 119.0 ±    14.6 5 0.017 1.08 ± 0.05 97.9 ±    13.6 6 0.020 0.95 ± 0.05 84.1 ±    14.5 7 0.025 0.77 ± 0.04 62.3 ±      2.9 8 0.030 0.67 ± 0.05 52.0 ±      2.7 9 0.040 0.51 ± 0.04 37.7 ±      2.0 10 0.050 0.41 ± 0.04 30.1 ±      1.8 11 0.060 0.34 ± 0.05 25.3 ±      1.6  40  Figure 2.1: Schematic of Phantom A.  Representative design of Phantom A showing all 11 tubes in a glass container fully submerged in water. While Tube 1 consisted of pure water, Tubes 2 to 11 contained increasing concentrations of MnSO4.H2O. Drawing not to scale. The consecutively designed Phantom B was based on the phantom previously investigated by Meyers et al. (65) for validation of the T2 relaxation based water content mapping method with a 32-echo GRASE sequence. Minor modifications to meet minimum geometric requirements for MRS measurements included the use of four narrow mouth 60 mL polypropylene laboratory bottles (VWR® International, LLC) as metabolite phantoms. The four different samples were designed to model water content, [H2O], relaxation behaviour and metabolite concentrations of NAA within a variety of different tissue types in vivo. Only one metabolite was used to reduce variability introduced by the spectroscopy acquisition as well as fitting procedure. Variation of the MR visible water content from 69 % to 83 % was implemented through the use of heavy water (deuterium oxide (D2O)), in which the 1H-atoms are replaced with the rarer isotope of heavy hydrogen, 2H. This isotope has a different gyromagnetic ratio and is thus not MR visible Test Tube 1Test Tubes 2-1141 with RF hardware tuned for a resonance frequency of 128 MHz for 1H at 3T. The range of water content aimed to replicate in vivo water content of different tissue types (66). The samples were prepared using four specific solutions, with the final composition of the sample only differing by the used volumes of each solution. The solutions used included (1) 50 mmol/L NAA dissolved in water and pH buffered (pH = 4.5), (2) MnSO4.H2O dissolved in d-H2O with a concentration of 0.663 mmol/L, (3) 99 % D2O (Sigma-Aldrich®) and (4) pure, de-ionised water, d-H2O. The pH of all samples was measured before sealing each bottle with laboratory film. The final composition of each sample solution is shown in Table 2.2. For measurements in the MRI scanner, the bottles were individually placed in a 1600 mL glass container filled with fresh tap water and were held in place using butcher’s yarn and adhesive tape to fixate the bottle in the centre of the container in all three dimensions. A schematic drawing of Phantom B is shown in Figure 2.2. Table 2.2: Prepared sample concentrations for in vitro phantom validations of Phantom B. To limit variability in relaxation times, all samples were prepared with the same concentration of manganese sulfate. Concentrations of the sample metabolite NAA and water content were varied between samples modelling a range of concentrations typically present in vivo. Sample # [NAA] (mmol/L) [H2O] (%) [D2O] (%) [MnSO4.H2O] (mmol/L) 1 12.0 73.0 27.0 0.118 2 8.0 69.0 31.0 0.118 3 12.0 83.0 17.0 0.118 4 10.0 76.0 24.0 0.118  42  Figure 2.2: Schematic of Phantom B. Representative design of Phantom B showing an individual Sample X in a glass container fully submerged in water. The polypropylene bottles are loosely held in place with yarn and adhesive tape at the centre of the 1600 mL container. Drawing not to scale. 2.2.2 MR Experiments 2.2.2.1 Phantom A: Relaxation Measurements  To measure the effect of different concentrations of MnSO4 on water T1 and T2 relaxation times, the phantom was scanned on a 3T Philips Achieva MR system (Best, The Netherlands) with an 8-channel SENSitivity Encoded (SENSE) head coil. The MR protocol included a sagittal 3D-T1 weighted MPRAGE acquisition with TE/TR/TI = 3.5 ms/8.1 ms/1052 ms, FOV(ap,rl,fh) = 256 mm/140 mm/256 mm and a reconstructed resolution of Δx/Δy/Δz = 1/1/1 mm³ for segmentation purposes. T1-mapping was performed with a 3D-Inversion Recovery (IR) sequence with low flip angle excitation (12°) complemented with a 3D ultra-fast gradient echo read-out with a 256x256 matrix (FOV=160 mm/132 mm/160 mm) and a reconstructed resolution of Δx/Δy/Δz = 0.9 mm /0.9 mm/2.0 mm). Hyperbolic secant pulses Fixation with yarn and tape in centre of container60 mL43 were applied to invert the magnetisation with inversion delays of 150 ms, 400 ms, 409 ms, 674 ms, 750 ms, 1113 ms, 1200 ms, 1836 ms, 2100 ms, 3030 ms and 5000 ms before the above-described excitation and readout (69). The shot interval (real TR) was set to the maximum allowed value of 6s, to allow for the magnetisation to return to equilibrium. For T2-mapping, a combined gradient and spin-echo (GRASE) sequence was applied with 48 echoes, shortest TE and echo spacing of 8 ms and TR of 1073 ms (FOV= 230 mm /190 mm /50 mm, matrix size = 240, slice oversampling factor = 1.3, Δx/Δy/Δz (reconstructed)= 1 mm /1 mm/2.5 mm) (67). 2.2.2.2 Phantom B: Validations Each of the four samples which comprised Phantom B was scanned with an identical hardware set-up as described for Phantom A, and similar conventional and quantitative imaging sequences. The acquisition protocol was supplemented with a SENSE reference scan to map coil sensitivities, a short-TE, single-voxel PRESS acquisition to measure NAA concentrations, and a shorter 5 min IR-T1 mapping series with shot-intervals of 3000 ms and TIs of 150 ms, 400 ms, 750 ms, 1200 ms & 2100 ms. An additional IR-T1 mapping series with a shot interval of 6000 ms and eight logarithmically spaced inversion times ranging from 150 ms to 5 s times was also acquired. The FOV was increased to 230 mm /190 mm /100mm to accommodate the larger dimensions of the bottles and achieve full coverage of the spectroscopic VOI within the metabolite sample. Critical parameters of the full protocol are listed in Table 2.3 and Table 2.4. 44 Table 2.3: Imaging sequence parameters for measurements in Phantom B.  SEQUENCE  SENSE-REFERENCE 3D-T1w MPRAGE GRASE IR-T1 SERIES SHORT IR-T1 SERIES LONG CONTRAST PARAMETERS TE / TR, (ms)  0.74 / 4.0 3.5 / 8.1  8, 16, … , 384 / 1073 4.6 / 8.0 4.6 / 8.0 TIs, (ms)  1052  150, 400, 750, 1200, 2100 150, 409, 674, 1113, 1836, 3030, 5000 Shot Interval, (ms)  3000  3000 6000 Flip Angle, (°) 1 8 90 12 12 Readout Bandwidth, (Hz) 2071.3 191.5 190 216.8 216.8 Acceleration Factor  - 250 (TFE)  3 (EPI) 100 (TFE) 100 (TFE) GEOMETRY PARAMETERS Acquisition Matrix (M×P) 96 × 75 256 × 250 232 × 93 232 × 190 232 x 190 Recon Matrix 96 256 240 240 240 FOV (ap/lr/fh), (mm) 300 / 530 / 530 256 / 140 / 256 230 / 190 / 100 230 / 190 /100 230 / 190 / 50 Voxel size:  Acquired (ap/lr/fh), (mm) 6 / 7 / 5.5  1 / 1 / 1 1 / 2 / 5 1 / 1 / 5 1 / 1 / 5 Voxel size: Reconstructed (ap/lr/fh), (mm) 3 / 7 / 5.5 1 / 1 / 1 1 / 1 / 2.5 1 / 1 / 2.5 1 / 1 / 2.5 SENSE factors - 1.8 (RL) 2 (RL) 1.5 (P-RL) 1.2 (P-os) 1.5 (P-RL) 1.2 (P-os) Readout 3D – FFE (cartesian) 3D – FFE (cartesian) 3D – SE (cartesian) 3D – FFE (radial) 3D – FFE (radial) Slice Orientation coronal sagittal axial axial axial Scan Duration, (m:ss) 0:44 5:26 7:31 5 × 1:17 = 6:25  8 x 1:12 = 9:36  45 Table 2.4: MR Spectroscopy sequence parameters for four samples of Phantom B. SEQUENCE single voxel PRESS TE / TR / NSA 31 ms / 4000 ms / 64 Phase Cycle Steps 16 Flip Angle, (°) 90 B1 rms, (µT) 0.45 GEOMETRY VOI Orientation coronal VOI Dimensions (ap/rl/fh), (cm) 1.3 / 1.3 / 4.0 = 6.8 mL  Plan Scan Metabolite / Centre Frequency NAA Shifted Metabolite Display H2O Chemical Shift Directions (ap/rl/fh) A / R / F PRESCAN PARAMETERS Water Suppression Method Excitation Water Suppression Window, (Hz)  200 # Water Reference Acquisitions 16 Shim routine 2nd order pencil-beam Shim Size (ap/rl/fh), (cm) 2.1 / 1.9 / 4.7 Scan Duration (m:ss, approx.) 5:28 2.2.3 Image Analysis 2.2.3.1 Quantitative Relaxation and Water Content Mapping IR-T1 data from Phantoms A and B were fit in MATLAB (R2018b, The MathWorks®, Inc.) with a least-squares approach. Voxelwise modelling of a single exponential recovery was applied to the magnitude images with varying inversion delay times to calculate quantitative T1-maps. Multi-echo GRASE data were fit with the regularised non-negative least squares (NNLS) with extended phase-graph stimulated echo correction with 48 logarithmically spaced T2 bins 46 ranging from 15 ms to 5 s and an assumed T1 of 1s (70–73). Fitting parameters for the T2-distributions were modified from 40 to 48 bins and 15-2000ms to 15-5000ms to better characterize the fraction of water with a long T2 in the bulk water. Water content mapping for Phantom B was performed as described previously by S Meyers using the reference scan method for correction of B1 receiver inhomogeneity (64, 65). Spectroscopy water relaxation maps were calculated voxelwise within the metabolite sample ROIs. The global geometric mean T2, obtained as part of the NNLS fitting routine, served as a surrogate for a hypothetical mono-exponential decay of water and was combined with the T1-map to calculate voxelwise spectroscopic water relaxation maps within the metabolite sample ROIs. 2.2.3.2 MR Spectroscopy Analysis Single-voxel proton spectra for each sample in Phantom B were eddy-current corrected during post-processing on the scanner’s platform. Each metabolite spectrum was fit in LCModel (Version 6.3 1-H) with the water scaling option (42). S Provencher provided a vendor and TE-specific metabolite basis-set with 16 metabolites simulated in GAMMA (74) for fitting. NAA concentrations calculated by LCModel in reference to the unsuppressed water peak were scaled with sample-specific water concentrations and relaxation attenuation by mapping the location of the voxel onto the quantitative maps and computing the mean water content, WC, and relaxation value, RATTH2O,  within the spectroscopic VOI. The metabolite concentration was not corrected for signal attenuation of the metabolite signal due to the relatively short TE of 47 30 ms and relatively long TR of 4s. The final absolute metabolite concentration of NAA in mmol/L was given by Equation (2.1).  [𝑁𝐴𝐴] (mmolL) = 𝑐 (𝑁𝐴𝐴𝐻2𝑂)⏟    𝐿𝐶𝑀𝑜𝑑𝑒𝑙 𝑅𝑒𝑠𝑢𝑙𝑡⋅ [𝐻2𝑂] (%) ⋅ 55.556mmolL⏟        𝑀𝑜𝑙𝑎𝑟𝑖𝑡𝑦𝑜𝑓 𝑃𝑢𝑟𝑒 𝑊𝑎𝑡𝑒𝑟⋅ 𝑅𝐴𝑇𝑇𝐻2𝑂 (2.1) 2.2.3.3 Structural Analysis For Phantom A, the IR-T1 images of the longest inversion delay were used to semi-automatically segment the individual tubes in FSLEyes (Version 0.27.3) using the intensity-based seeding tool to identify the 11 separate tubes (75). This segmentation was applied to the quantitative T1 and T2 maps to mask the individual tubes. The mean and standard deviations of each tube’s relaxation constant were calculated.  Each of the four individual phantoms scanned for Phantom B was processed identically. The first echo image (TE = 8ms) of the GRASE sequence was semi-automatically segmented with tools in FSLEyes (details as above) into two regions of interest (ROI); the first ROI was comprising voxels within the metabolite samples, and the second ROI comprising imaging voxels in the surrounding pure water.  The body coil portion of the SENSE reference scan for Phantom B was manually registered to the first echo of the GRASE sequence in 3D Slicer (Version 4.10.0) with the Transform and Resample Image modules to correct for receiver B1 inhomogeneities (76). 48 2.2.3.4 Statistical Analysis Two linear regression models were fit to the measurements of Phantom B, with the measured water contents [H2O] and NAA concentrations being the dependent variables and using the prepared [H2O] and NAA concentrations as independent variables, respectively. Statistical analysis was performed in MATLAB (R2019a, The MathWorks®, Inc) with the fitlm function. 2.3 Results 2.3.1 Determining a Concentration for a T1 & T2 Shortening Agent The results for the relaxation times of the different concentrations of MnSO4 in Phantom A are summarised in Table 2.1 and visualised in Figure 2.3. A total of four test-tubes with MnSO4 concentrations between 0.013 and 0.025 mg/ml replicated relaxation values which may be expected in healthy brain tissue (as indicated by dashed horizontal lines) with measured T1 values ranging from 0.77 ± 0.04 s to 1.23 ± 0.05 s and T2’s from 62.3 ± 2.9 ms to 119.0 ± 14.6 ms. The segmentation approach for in vivo scaling of metabolite concentrations typically uses T2 correction constants of approximately 70 ms. This behaviour is most closely modelled by sample tube #6 with a concentration of 0.020 mg/mL (T2 = 84.1 ± 14.5 ms) and still maintained a spin-lattice relaxation time of approximately 1s. These observations implied the use of a concentration of 0.020 mg/mL MnSO4 (the equivalent of 0.118 mmol/L) for the sample preparations of Phantom B. 49  Figure 2.3: Measured water relaxation constants of Phantom A for variable concentrations of manganese sulfate monohydrate in deionized water. Orange data points indicate the measured mean water relaxation constants T1 (left) and T2 (right) for each sample with the error bars indicating the standard deviation. The dark blue dashed horizontal lines represent the range of typical in vivo relaxation times while the light blue dashed constants are the lower and upper bound of relaxation constants which were observed by Meyers et al. in their phantom validation studies (65). 2.3.2 Water Content Mapping with a 48-echo GRASE Sequence The measured water contents [H2O] in per cent are plotted against the target concentrations, which were prepared in the laboratory in Figure 2.4 . The prepared water contents were an excellent predictor for the measured water contents (R²= 0.985, root mean square error (RMSE) = 0.69, p = 0.00743) while the bias, indicated by the y-intercept of the linear regression, y0 = 21.6%, was not significant (p = 0.051) at a significance level of α = 0.05. Overall, water content within the sample bottles was overestimated by 4.8 % on average. 0 0.1 0.2 0.3102103   Depen ence o  n    MnSO4  (mmol/L)T2 of H2O (ms)Fit Results: R2 = 1.000, RMSE = 3.2690 0.1 0.2 0.300.511.522.533.5   Depen ence o  n    MnSO4  (mmol/L)T1 of H2O (s)in-vivo scaling WMin-vivo scaling  MFit Results: R2 = 0.999, RMSE = 0.02850  Figure 2.4: Accuracy of water content, [H2O], measurements in vitro. Prepared and measured water contents for samples 1 through 4 of Phantom B. Error bars reflect the standard deviation of water content values within each bottle.  Across all samples, the measured water contents were overestimated by 4.8% on average, in comparison to the prepared water contents. 2.3.3 Accuracy of In Vivo Metabolite Concentrations Phantom spectra were of high quality with a high median SNR of 44.5 (range: 34 – 48), and narrow LCModel determined FWHM of 1.4 Hz (range: 1.4-1.9 Hz). Figure 2.5shows a representative sample spectrum and LCModel fit. The measured [NAA], as well as the linear regression results, are shown in Figure 2.6. For all four samples measured [NAA] was underestimated by 9.9% on average. 51  Figure 2.5: Representative spectrum from Sample X for Phantom B.  With an SNR of 48 and FWHM of 1.4 Hz NAA was fit with a Cramer-Rao Lower Bound (CRLB) as low as 2 %. A flat baseline and randomly distributed residuals are further indicators of high spectral quality.  52  Figure 2.6: Accuracy of NAA concentrations, [NAA], measurements in vitro. [NAA] as prepared (x-axis) and measured (y-axis) with the proposed water scaling method. Note that the measured concentrations are lower than the prepared ones (pairwise mean difference = - 9.9 %). The linear regression analysis predicts a significant correlation (Adjusted R² = 0.885, RMSE = 0.51, p = 0.039), while the bias of y0 = 1.49 mmol/L, represented by the y-intercept, was not deemed significant (p = 0.46). 2.4 Discussion and Limitations Relaxation constants were measured for different concentrations of the T1 and T2 shortening contrast agent MnSO4.H2O and provided a range of concentrations with corresponding relaxation times as might be present in the human CNS at 3 T (77, 78). The relaxation measurements provided essential knowledge for the design of Phantom B. 53 2.4.1 Sample Preparation The preparation of samples for Phantom B was carefully planned and executed, aiming for highly precise and accurate sample preparations. Unfortunately, some factors were beyond the researcher’s control when preparing the sample solutions. For instance, the metabolite solution was assembled with an NAA solution prepared by previous researchers. While a concentration of 50 mmol/L, as indicated by the label, was assumed for the sample preparation, the exact metabolite concentration with error estimates is unknown. Furthermore, the pH of approximately 7.45 indicates other components are present in the 50 mmol/L solution since NAA is acidic. It may be assumed that a chemical compound, e.g. sodium phosphate monobasic monohydrate, was used to buffer the pH at around 7.4 ± 0.1. This is further suggested by the pH analysis of the final sample solutions prepared for Phantom B, which despite having varying concentrations of NAA and the acidic MnSO4.H2O, revealed that pH was stable in a range from 7.38 to 7.45.  Possible imprecision in the sample preparation could also cause the variability in relaxation constants of the samples. Given that each of the four samples should have identical concentrations of MnSO4.H2O and therefore relaxation behaviour, it is interesting to note that T1 and T2 relaxation values ranged from 0.634 s to 0.764 s for T1 and 139ms to 157ms for T2. This indicates a sensitivity to differences in concentration of the contrast agent. It can also be explained by previous reproducibility studies which used similar sequences and reported low scan-rescan coefficients of variations. This can also contribute to the observed variability in relaxation constants (77, 79).  54 These uncertainties indicate necessary improvements in techniques used for sample preparations for future phantom investigations. Possible approaches would include using elemental NAA with known molar weight and preparing metabolite solutions with this compound in order to estimate measurement uncertainties.  2.4.2 Phantom Design & Temperature Samples for Phantom B were placed in a larger glass container filled with pure tap water. To restrict motion of the sample bottles, they were held in place by several strings fixated at the bottom of the container. It is, however, difficult to estimate how mobile the sample, as well as the surrounding water, was. MR images did not reveal any gross motion of the sample, but gradient intense sequences, e.g. GRASE imaging, are prone to make the MR tabletop and samples vibrate. In our phantoms, this was observable in the first GRASE echo, showing evidence of circular ripple artefacts in the surrounding water. It was also indicated by the sum of squares of the residuals quality criterion for identifying voxels prone to flow for the TWC mapping technique. Sandbags placed on top of the receive coil reduced this artefact qualitatively but were not able to eliminate it completely. One may want to consider embedding the sample in a more viscous or even gel-like environment, thus minimising mechanical vibrations. A popular and simple option for this is the use of agar as a gelling agent (80–82). Embedding the metabolite sample bottle in this environment and adding a second sample tube or bottle containing pure water would likely reduce mechanical vibrations. A further point to consider in the interpretation of the results presented is the impact of temperature on the MR signal. In Equation (1.3) we showed that the magnetisation and thus, 55 MR visible proton density is inversely proportional to the temperature. For instance, the difference in magnetisation between a sample at room temperature of 21°C and a sample with cold water, e.g. approx. 15°C would be 2.1%. This is of importance since the metabolite samples were prepared at room temperature, but then embedded in fresh tap water of a lower temperature only shortly before the MR scan was performed. It is thus highly likely that the metabolite samples and water reference standard had different temperatures at the time of measurement. This could be omitted by preparing the phantom in advance and allowing it to adapt to the surroundings by placing it in the scan room 12 to 24 hours in advance. 2.4.3 MR Measurements Temperature is also a parameter to consider for the MR measurements because it impacts the relaxation behaviour of water. In particular, T1 is readily affected by changes in temperature (83). For water, T1 decreases with temperature. When measuring relaxation times, it is preferable to sample inversion times up to approximately thrice the expected relaxation constant. For this reason, the longest shot interval of 6s with inversion times of up 5s was used measuring the T1 in the pure water standard, but a shorter inversion recovery series was used for T1 measurements within the metabolite sample. Given the measured relaxation times ranged from 2.4s to 2.8s for the four samples, it would be beneficial to use a T1 mapping sequence optimised for very long T1 times. Further increasing the shot interval may however only be of limited applicability, since scan time will increase proportionally with it. Other T1 mapping approaches, e.g. variable flip angle techniques, could prove useful in future studies to address time restrictions (84). 56 2.4.4 Accuracy of Water Content Measurements Using a 48 echo GRASE Sequence.  Water content was overestimated by 4.8% on average in comparison to the prepared water contents, which was larger than previous phantom measurements with the 32-echo GRASE sequence which reported accuracies of approximately 2% (65). It can partially be explained by potential temperature differences between the sample and surrounding water and potential inaccuracies in T1 measurements. The calculation of water content furthermore involves image masks and registrations, which are automated for in vivo analyses. For this phantom study, both segmentation of the metabolite sample and water standard were conducted semi-automatically, introducing a possible reader dependence. Not only were segmentations done manually, but since the coil sensitivity reference scan used to correct for receiver B1 inhomogeneities offers only minimal contrast, an automated registration was not successful and manual registrations had to be performed. For future investigations, it would be preferable if these analysis steps could be automated to eliminate reader-variability. 2.4.5 Accuracy of NAA Concentrations Metabolite concentrations of NAA were underestimated by 9.9 % on average in comparison to the prepared concentration. While it is possible that the assumed concentration of 50 mmol/L for the solution of NAA may lead to incorrectly prepared concentrations, it cannot fully explain the difference in measured and prepared concentrations. Furthermore, absolute metabolite concentrations in theory scale proportionally with the assumed water concentration. Since 57 water concentrations were overestimated, it is surprising that this does trend was not carried forward when calculating absolute metabolite concentrations of NAA. It challenges the assumption of a negligible metabolite signal attenuation for the sequence and sample used. It may be that when measuring the concentration samples for Phantom B at a TE of 30 ms and TR of 4 s, metabolic signal attenuation is of importance and non-negligible. The impact of the used relaxation shortening agent MnSO4 on the relaxation of metabolites including NAA has only been studied for T1 (85) and demonstrated that at a pH of 7.4 the relaxivity r1 of Mn2+ ions for the NAA singlet (r1,Mn(NAA) = 28.1 s-1 mMol/L-1) is three times stronger than for water (r1,Mn(H2O) = 8.3 s-1 mMol/L-1). This strong interaction is, in a simplified model, caused by the oppositely charged acidic NAA2- molecule and Mn2+ ions. While no data on the impact of Mn2+ on the transverse relaxation of NAA was found, the literature on the impact of negatively charged gadolinium chelates and positively charged choline molecules (86), suggest a similar relationship. This could cause a considerably shorter T2 for the NAA singlet, and consecutively a stronger metabolite signal attenuation, warranting a correction. To investigate the impact of Mn2+ on the NAA relaxation further, one would want to acquire multiple-echo and multiple inversion recovery spectroscopy data to measure the T2 and T1 of the NAA peak at 2.02 ppm, respectively. 2.5 Conclusion We measured the relaxation behaviour of water for varying concentrations of the relaxation shortening compound MnSO4. At a concentration of 0.020 mg/mL, water relaxations were within the biologically observed range and could be used to prepare metabolite solutions 58 containing varying concentrations of NAA and MR visible water content. While water content was overestimated by approximately 4.8%, measured NAA concentrations were 9.9% lower than prepared. These inaccuracies should be investigated in future phantom studies with more than four sample concentrations and an improved phantom design and measurement procedure, as outlined in section 2.4. Overall, the proposed technique shows promise to be used in vivo where challenges present in vitro, e.g. temperature variations as well as mechanical vibrations are not of a concern for healthy controls. 59 Chapter 3: Towards Personalised Water Scaling of In Vivo Metabolite Concentrations. 3.1 Introduction In human studies, MRS data is often expressed as a ratio of two concentrations of metabolites, e.g. NAA/tCr. These relative concentration measurements, despite being sensitive to changes in certain scenarios, are not specific and can thus lead to misinterpretation of the underlying physiological changes (53). Furthermore, the non-specificity of metabolite ratios complicates cross-sectional or longitudinal comparisons in clinical and pre-clinical studies. This dilemma popularised the approach of referencing metabolite concentrations with respect to other internal standards e.g. tissue water. Estimation of tissue water is easily implementable by acquiring high-resolution anatomical sequences and using partial tissue volume contributions to the spectroscopic VOI as scaling factors for constant tissue specific water properties. A key parameter obtained through tissue segmentation, which is used in calculating metabolite concentrations, is determining the volumetric contribution of CSF within the VOI, given that there are no metabolites in CSF. Fitted metabolites concentrations are scaled with a factor of 𝑘 =11−𝑓𝐶𝑆𝐹 to calculate absolute tissue metabolite concentrations in reference to the water signal. Thus, an accurate estimate of the signal originating from CSF in the VOI is critical. Inconsistent CSF contributions between subjects can introduce variability in the calculated absolute metabolite concentrations, either masking a change or providing evidence of changes in metabolite concentrations, when it is, in fact, the CSF signal fraction that is changing. For a biologically relevant range from 0 % to 30 % CSF fraction, a simple calculation reveals, that the 60 absolute metabolite concentrations scale with the percentage change in CSF fraction. Thus, the scaling of the metabolites’ signals could vary by up to 30% if inaccurate CSF fractions are assumed. Therefore, it is critical to accurately determine the fraction of CSF present in the voxel. This can either be obtained from high-resolution anatomical images, e.g. a 3D-T1 weighted image (fCSF,V)or by exploiting the differences in T2 relaxation between bulk and tissue water. The volumetric fraction of CSF, fCSF,V, can be estimated by segmenting high resolution anatomical images, or alternatively, may be calculated with the aid of multi-echo T2 data, fCSF,S. Multi-echo T2 data enables separation of the signal fraction originating from tissue water (short and medium long T2s shorter than 200 ms) and the signal contribution by CSF which, since it is primarily free bulk water, exhibits T2s of typically at least 600ms at 3T in healthy controls (64). Other tissue specific constants, including water content and relaxation properties, could change in disease, which would not be identified by tissue segmentation. Thus, using literature values may not always be the most accurate option. Therefore, following the investigation of the proposed water content mapping and water signal relaxation correction method in phantoms, the aim of this chapter was to determine how feasible the technique is in comparison to frequently used water scaling techniques in vivo (87, 88). 3.2 Methods 3.2.1 Sample Overview and Experimental Set Up Data was collected on a 3T Philips Achieva (Best, The Netherlands), release 3.2.3) MR system. The quadrature body coil was used for RF transmission and an 8 channel Sensitivity Encoding 61 (SENSE) head coil was used for signal reception. All subjects were scanned in the supine position, head first. Two layers of hearing protection (earplugs and headphones) were provided, and the subjects’ heads were immobilised with a band across the forehead and foam cushioning inside the head coil to prevent motion. 3.2.2  Main Study 3.2.2.1 Participant Information Consent for this study was obtained from ten healthy young adults with no known neurological conditions under the Magnetic Resonance Measurements on Normal Volunteers Ethics. The volunteers (mean age (range) = 23.4 (21-26) years, 5 female, 5 male) were scanned between September 29th, 2018 and January 11th, 2019. 3.2.2.2 Scan Protocol The main scan protocol consisted of seven MR acquisitions: 1. T1 weighted localizer survey (TE/TR/TI=4.6 ms, /11 ms/ 800 ms, FOV (ap/rl/fh) = 250mm/180mm/250mm, reconstructed voxel size = 1 mm / 1 mm / 10 mm), 2. Reference scan to determine the coil sensitivity maps of the transmit and receive coils (Table 3.1) 3. T1-weighted 1 mm³ isotropic MPRAGE sequence for planning of subsequent MRS voxels (Table 3.1) 4. central white matter (cWM) single voxel spectroscopy (SVS) with (PRESS, TE=31ms, TR=4s, single voxel in left central white matter (cWM), size (ap/rl/fh) = 6.5 cm / 1.5 cm/ 2.0 cm = 19.5 mL) sequence) (36). A 16-step phase-cycle scheme, water suppression 62 through Excitation and automated second-order shimming with the pencil-beam technique were applied. The synthesiser frequency was centred on the NAA singlet, with the chemical shift directions of the VOI for water facing away from the lateral ventricles. The volume for shim optimisation was manually prescribed to include both the NAA singlet as well as the shifted metabolite display for water. No inner- or outer volume signal saturation was used in this sequence. The chemical shift displacement was 1.4 mm/ppm in the l-r direction, 3.5 mm/ppm in the a-p direction and 1.8 mm/ppm in the f-h- direction as determined with the shifted metabolite display on the scanner User Interface. 64 individual metabolite spectra and 16 water reference spectra for spectral corrections were acquired, leading to a total scan time of 7:20 per VOI. This included preparatory phases, automated shim and water suppression optimisation and 80 individual SVS acquisitions. For offline post-processing, each of the 80 individual spectra were exported (Table 3.2). The water content and relaxation attenuation mapping protocol included a 5. 48-echo 3D-GRASE sequence (27) with shortest TE and TE spacing of 8ms (Table 3.1),  6.– 10. A series of five inversion recovery sequences (69). For a shot interval (actual TR) of 3 seconds, the signal for inversion times of 150, 400, 750, 1200 & 2100 ms was measured. This series is hereafter referred to as short-TR (shot interval = 3s) series (Table 3.1). 11. A second anatomical scan (11) with T2-weighting followed the IR-T1 series to identify potential subject movement (TE/TR=363ms/2500ms, FOV(ap/rl/fh)=252 mm /165 mm /256 mm, reconstructed voxel size = 1 mm/ 1 mm/ 1 mm, TSE factor 120).  63 12. After confirming the absence of substantial subject motion, a second short-TE MRS acquisition was conducted in the interhemispheric parietal grey matter, with similar sequence parameters as acquisition (4) and adjusted voxel geometry to suit the anatomy (VOI dimension (ap/rl/fh) = 3.0 cm / 2.5 cm / 2.2 cm, Table 3.2). An extended protocol was conducted in 4 of the 10 subjects to acquire an improved T1 measurement of CSF. This protocol appended an IR-T1 series after the second SVS acquisition, in which the shot interval was set to the 6s, the maximum allowed value, with a logarithmic spacing of the inversion times (TIs = 150, 248, 409, 674, 1113, 1836, 3030, 5000 ms). The FOV and slice stack were decreased to compensate for the increased scan time due to elongated TRs. This series is hereafter referred to as long-TR (shot interval = 6s) series.  Essential acquisition parameters of key imaging and spectroscopy sequences are summarised in  Table 3.1 and Table 3.2, respectively. 64 Table 3.1: MR Imaging sequence parameters  SEQUENCE  SENSE Reference 3D-T1W MPRAGE GRASE IR-T1 Series CONTRAST PARAMETERS TE/TR, (ms) 0.74 / 4.0 / - 3.5 / 8.1 8,16,…,384/1073/ -  4.6 / 8.0 TI, (ms) - 1052 - 150, 400, 750, 1200, 2100 Shot interval, (ms) - 3000 - 3000 Flip Angle, (°) 1 8 90 12 Readout Bandwidth, (Hz) 2071.3 191.5 190 216.8 Acceleration Factor  - 250 (TFE) 3 (EPI) 100 (TFE) GEOMETRY PARAMETERS Acquisition Matrix (M × P) 96 × 75 256 × 250 232 × 93 232 × 190 Recon Matrix 96 256 240 240 FOV (ap/lr/fh), (mm) 300 x 530 x 530 256 / 165 / 256 230 / 190 / 100 230 / 190 /100 Voxel size:  acquired (ap/lr/fh), (mm) 6 / 7 / 5.5  1 / 1 / 1 1 / 2 / 5 1 / 1 / 5 Voxel size: reconstructed (ap/lr/fh), (mm) 3 / 7 / 5.5 1 / 1 / 1 1 / 1 / 2.5 1 / 1 / 2.5 SENSE factors - 1.8 (RL) 2 (RL) 1.5 (P-RL) 1.2 (P-os) Readout 3D – FFE (cartesian) 3D – FFE (cartesian) 3D – SE (cartesian) 3D – FFE (radial) Slice Orientation coronal sagittal axial axial Scan duration, (m:ss) 0:44 6:26 7:31 5 × 1:17 = 6:25  65 Table 3.2: MR Spectroscopy sequence parameters  central White Matter (cWM) parietal Grey Matter (pGM) SEQUENCE single voxel PRESS single voxel PRESS TE / TR / NSA 31 ms / 4000 ms / 64 31 ms / 4000 ms / 64 Phase Cycle Steps 16 16 Flip Angle, (°) 90 90 B1 rms, (µT) 0.45 0.45 GEOMETERY PARAMETERS VOI Orientation sagittal transverse VOI Dimensions (ap/rl/fh), (cm) 6.5 / 1.5 / 2.0 = 19.5 mL  3.0 / 2.5 / 2.2 = 16.5 mL  Plan Scan metabolite / centre frequency NAA NAA Shifted metabolite display H2O H2O Chemical Shift Directions (ap/rl/fh) A / R / F P / R / F PRESCAN PARAMETERS Water Suppression Method Excitation Excitation Water Suppression Window, (Hz)  160 160 # Water Reference Acquisitions 16 16 Shim routine 2nd order pencil-beam 2nd order pencil-beam Shim Size, (cm) 8.2 / 2.0 / 2.6  3.9 / 3.2 / 3.0  Approx. Scan Time, (m:ss)  7:20 7:20 66 3.2.3 CSF Sub-Study  An eleventh volunteer was scanned to investigate the sensitivity of the IR-T1 series to CSF flow in the lateral ventricles. This scan protocol consisted of  a 3D-T1 weighted MPRAGE anatomical sequence (see Table 3.1) and three inversion recovery series, each with eight inversion times (TIs = 150, 248, 409, 674, 1113, 1836, 3030, 5000ms) and a shot-interval of 6s. IR-T1 series #1 and #2 were identical to assess scan-rescan reliability, while for the third series the phase encoding direction (aka “fold-over direction”) was changed from right to left in Series 1 & 2 to anterior to posterior in Series 3. 3.2.4 MRI and MRS Data Processing 3.2.4.1 GRASE Analysis The 48-echo GRASE data was analysed with a regularised NNLS algorithm in MATLAB (R2018b, MathWorks Inc) with stimulated echo correction and flip angle estimation (71). Default fitting parameters for the T2-distributions were modified from 40 to 48 bins with the T2 range changing from previously 15-2000ms to 15-5000 ms. The increased number of bins was used to better characterize the fraction of CSF in each voxel (65). A further modification to the analysis included an additional quantitative map of the squared sum of residual errors (SoS) of the fitted T2-decay. Other quantitative maps computed included the global geometric mean T2 (GMT2, or single-component T2 of water over 15ms to 5000ms range), the global density (GDN, the integral under the T2 distribution), the myelin water fraction (MWF), defined as the signal fraction with T2s between 15 and 40 ms. The CSF signal fraction, fCSF,S, was also calculated and defined as the fraction of signal demonstrating T2s of greater than 600ms (57). 67 3.2.4.2 IR-T1 Series The short and long IR-T1 series were analysed with a least-squares fit implemented with the Levenberg-Marquardt algorithm in MATLAB (R2018b, MathWorks Inc). The model used to fit the magnitude MR images is given in equation (3.1), with M0, β and T1 being variables to be determined.  𝑆(𝑇𝐼) = |𝑀0 ⋅ (1 − 𝛽 ⋅ 𝑒−𝑇𝐼𝑇1) |  (3.1) 3.2.4.3 Structural and Volumetric Analysis All images were brain-extracted with FSL’s brain extraction tool (BET, robust option, threshold 0.5) and segmented using FSL’s FAST algorithm to identify  M, WM and CSF in the partial volume estimates (FSL Version 6.0.1) (89–91). The first GRASE echo (TE = 8ms) and the 3DT2-weighted brain-extracted images were rigidly registered to the 3D-T1 weighted scan (FLIRT, 6 degrees of freedom (DoF), trilinear interpolation), and the inverse GRASE transformation matrix was applied to the 3DT1 to perform automatic segmentation of the lateral ventricles with ALVIN (92) in SPM8 (93). Each IR-T1 series with shot intervals of 3 s and 6 s were visually inspected to identify potential subject motion within each series of 5 and 8 images, respectively. To minimise misalignment between quantitative T1-maps and GRASE-derived T2 metrics, images with the greatest T1 weighting, TI (short series) = 1200 ms, TI (long series) = 1836 ms were first brain extracted and then rigidly registered to the first GRASE echo (FLIRT, 6 DoF). Due to a small FOV in the foot-head direction the long IR-T1 series was brain extracted with the -Z option and a conservative fractional intensity threshold of 0.3, before registration to GRASE space. Registration matrices 68 determined between high T1-contrast IR-T1 images and GRASE were applied to the quantitative T1-maps with tri-linear interpolation. To investigate the differences in CSF signal or volume contribution, the partial volume estimates for CSF, fCSF,V, obtained by FAST segmentation (91) (default parameters) of the 3D-T1w image were compared to the CSF signal fraction, fCSF,S, obtained through the NNLS fit of the GRASE data within the parietal GM VOI. For the T1 sub-study, the robust brain extracted 3D-T1 image was segmented with ALVIN to identify the lateral ventricles. Each IR-T1 series was visually inspected to identify potential subject motion within the series of eight scans. Due to a small FOV in the foot – head direction, the image with the best soft tissue contrast (TI = 1836 ms) was brain extracted with the -Z option and a conservative fractional intensity threshold of 0.3. The brain extracted inversion recovery images were then linearly registered to the 3D-T1 image (9 degrees of freedom, 360° search window, correlation ratio cost-function, sinc interpolation) and the registration matrix applied to the quantitative T1-maps from each series (tri-linear interpolation). 3.2.4.4 SVS Analysis Individual shots were zero-order phase corrected with the mean phase of the water peak, and a linear 1st order phase correction was applied to the metabolite spectra to compensate for eddy currents by applying the inverse phase of the water FID to the metabolite acquisitions (41). To compensate for slight changes in signal frequency, e.g. due to motion or a shift in centre frequency, each full phase cycle of 16 individual shots was then averaged so that 1 minute averages could be aligned in the frequency domain by applying a frequency shift in the time 69 domain. The frequency aligned metabolite spectra were then fitted as a single voxel spectroscopy scan with LCModel (version 6.3-1.H) in the frequency region from 0.2ppm to 4.0ppm (42). A vendor-specific- and TE-matched basis set with 16 metabolites simulated in GAMMA (74) was provided by S Provencher. The basis set did not include a measured macromolecular baseline. 3.2.4.5 Calculation of Water Scaling Factors (T1, T2, RATTH2O and TWC) Figure 1.9-B provides an overview of how the acquired data is used to calculate the necessary quantitative maps for calculating the personalised water scaling maps. In brief, the measurement of total water content utilises the concept that the equilibrium magnetisation, M0, is proportional to the number of hydrogen protons and thus water content. For the GRASE sequence, the signal at time 0 (S0) can be calculated by integrating over the T2-distribution which has been fitted to the decay curve. In the time-domain, this intercept is proportional to M0 and can be described expressed as  𝑆0 = 𝑘𝐵1−𝑀0 (1 − 𝑒−𝑇𝑅𝑒𝑓𝑓𝑇1 ) (3.2) after correcting for signal attenuation due to T2 decay and B1+ as part of the NNLS fitting procedure. Thus, dividing S0 voxelwise by a B1- map and the term (1 − 𝑒−𝑇𝑅𝑒𝑓𝑓𝑇1 ) one can obtain an expression in which the signal is proportional to M0, with the constant of proportionality being k, a constant which combines previously introduced parameters, e.g. receiver gain and other scaling parameters. Since this factor is constant across all voxels, one can eliminate it by referencing S0 in a specific voxel to S0 of a voxel with known water concentration, e.g. an external phantom with pre-determined water concentration or the CSF in the ventricles 70 (typically ≈ 99%), thus obtaining the desired water content. Correcting for the receiver inhomogeneity, B1- was achieved by applying the principle of reciprocity to a reference scan, required for SENSE reconstructions (94), and allowed the voxelwise calculation of water content maps. Attenuation of the water peak for the PRESS sequence with parameters as listed in Table 3.2 was also calculated per image voxel as per equation (1.13). Both, the effective [H2O] and RATTH2O were then defined and computed as the weighted average of the voxelmask with the individual TWC and RATTH2O maps. LCModel obtained metabolite concentrations scaled to the water peak were subsequently scaled with the effective RATTH2O and [H2O] factors and compared to the traditional, segmentation-based approach. 3.2.4.6 Quality Assurance and Statistical Analysis Each spectrum fit in LCModel was visually inspected by myself and a second independent spectroscopist with more than 10 years’ experience (ELM). The fit of a specific metabolite for each spectrum was deemed sufficiently reliable if the absolute Cramer-Rao-Lower Bound (CRLB) was less than 30% of the median absolute metabolite concentration across all ten volunteers (95). The relatively small sample size of ten volunteers warranted the use of non-parametric statistic analyses, including a Wilcoxon signed-rank test (96) to compare the water scaling factors, as well as the absolute metabolite concentrations, between the segmentation and the total water content approach. Additionally, a Brown-Forsythe-Test for equal variance was conducted to evaluate differences in the spread of the two quantification methods (97). For correlations 71 between the two CSF fraction measurements, a Spearman’s rank correlation was calculated (98). All statistical analyses were performed in MATLAB R2019a with the Statistics and Machine Learning Toolbox. Statistical significance for all comparisons were defined as p < 0.05. 3.3 Results 3.3.1 CSF Flow Effect on T1 Relaxation Three sample slices of the T1 maps resulting from the three separate IR-T1 series is shown in Figure 3.1. A qualitative comparison of these slices suggests that the first and third series have more similar values within the lateral ventricles than the first and second IR-series with identical scan parameters. This is confirmed when investigating the distribution of T1 values present within the lateral ventricles (see Figure 3.2). 72  Figure 3.1: Scan-rescan reproducibility of T1 maps of CSF (left, middle) and impact of phase encoding directions on T1 measurements of CSF in single individual (left, right). Three sample slices including the top, middle and lowest slice of the lateral ventricles. Note that the left and right T1 maps qualitatively show more similar values than the scan-rescan pair of the first and second series suggesting the used T1 mapping sequence is not noticeably affected by the direction of flow of CSF. T1 (s)Scan 1(phase: right-left)Scan 2(phase: right-left)Scan 3(phase: anterior-posterior)73   Figure 3.2: Histograms of ventricular T1 values. Note that Series 1 and 3 (B) qualitatively show higher similarity than the two scan-rescan measurements from Series 1 and 2 (A). This provides evidence that the T1-mapping sequence is not noticeably affected by the systematic flow of CSF, since this would affect the T1 distribution of Series 3 in a greater way than the distribution of Series 2. 3.3.2 TR Effects on T1 Mapping in CSF To accurately measure the T1 of CSF in the lateral ventricles, four volunteers were scanned with the expanded protocol involving two separate IR-T1 series with different shot intervals as described in Section 3.2.2.2. These series are hereafter referred to as short-TR (shot interval = 3s), and long-TR series (shot interval = 6s) and are shown in Figure 3.3. The relaxation times measured with the long-TR protocol was mean T1,CSF ± SD = 3.9 ± 1.9s, while with the short-TR protocol was mean T1,CSF ± SD = 2.1 ± 0.5s, suggesting that inversion times and shot intervals 2 4 6 8 10   (s)Normalise   re uency2 4 6 8 10   (s)( A ) ( B )74 should be prolonged, to measure the T1 of CSF more accurately, since previous studies reported values between 4.1s and 4.3s (69, 99).  Figure 3.3: Histograms for four volunteers showing the effect of different shot intervals (blue = 6s, orange =3s) and inversion times on the measurement of T1 in CSF. Dashed lines indicate the mean T1,CSF measured in the lateral ventricles of each volunteer, while solid lines represent the median for the long and short IR-T1 series for each volunteer. 3.3.3 Correlation of Segmentation and T2 Based Estimates of Voxel CSF Volume Fraction We compared two methods to quantify the contribution of CSF in to the spectroscopic region in the inter-hemispheric parietal grey matter. This location suffers from high inter-subject variability in the CSF fraction. For the present cohort, segmentation-based CSF fractions, fCSF,V,  were as high as 30% and as low as 8%. A good correlation was found between fCSF,S estimated from the T2 distributions and fCSF,V (Spearman's 𝜌 = 0.90, p = 0.001) (Figure 3.4). The median CSF fraction for the T2 method was 0.064 (range = 0.039 - 0.136), while the median CSF fraction 0 2 4 6 8 10 12   (s)0 2 4 6 8 10 12   (s)      s       s( A )( C )( B )( D )75 for the segmentation-based method was found to be 0.199 (range = 0.082 – 0.275), showing a significant difference of 12.8% (p = 0.002, Figure 3.4).  Figure 3.4: Correlation of CSF contributions identified by segmentation (x-axis) or T2 time (y-axis) within the pGM VOI for each volunteer. The horizontal orange line is the median of fCSF,S = 0.064 and the black vertical line is the median of fCSF,V=0.199. Note the significantly lower CSF contributions in the T2 signal fraction by -12.8% (p = 0.002). 3.3.4 Spectral Data Quality All data acquired with the described protocol was of excellent quality and provided no evidence of considerable subject motion. Automated tissue segmentation and inter-modal registration of imaging data was able to achieve sufficiently high accuracy without the need for manual editing as confirmed by quality assurance images confirming spatial alignment and segmentation 76 accuracy. Spectra exhibited excellent linewidths of mean 4.8 Hz and a range of 4.0-6.8 Hz in cWM and comparable results in pGM (mean FWHM (range) = 4.7 (3.4-5.9) Hz). The minimum SNR measured was observed to be 36 and could go as high as 48 in white matter and 45 in grey matter (Table 3.3). Figure 3.5 shows the mean pre-processed spectral data for each VOI.  Table 3.3: Data quality and VOI composition consistency Quality Assurance Metric Central White Matter (cWM) Mean                SD               Range Parietal GM (pGM) Mean                SD               Range SNR 42.9 2.8 39.0 - 48.0 40.5 3.0 36.0 - 45.0 LCModel FWHM (Hz) 4.8 0.9 4.0 - 6.8 4.7 0.7 3.4 - 5.9 fCSF 0.02 0.01 0.01 – 0.03 0.17 0.06 0.05-0.25 fWM 0.86 0.03 0.80 – 0.90 0.26 0.05 0.21-0.35 fGM 0.12 0.02 0.09 - 0.17 0.57 0.04 0.51-0.61 77  Figure 3.5: Pre-processed spectra for both VOIs averaged across ten healthy controls (black). All spectra exhibit high quality with narrow linewidths and good consistency between subjects. Shaded grey are indicates the range of spectra of the ten subjects. Segmentation of the two VOIs furthermore revealed that tissue compositions of the voxels were highly consistent in their primary tissue component. For the cWM VOI, the voxel had a minimum white matter volume contribution of 0.80 (mean fWM = 0.86), a negligible CSF volume contribution of 0.03 or less, while the contribution of GM was consistently below 0.17 with a mean of 0.12. The interhemispheric pGM VOI constantly contained a majority of grey matter (mean=0.57, minimum contribution of 0.51), with more variable white matter (fWM =0.26 ± 0.05) 0.511.522.533.54Chemical  hi  (ppm)( A ) Central hite  a er( B )  arietal Grey a er78 and CSF contributions (fCSF = 0.17 ± 0.06), due to a more heterogeneous environment within the interhemispheric VOI. The excellent spectral quality permitted the reliable fit of the five major metabolites of NAA, tCr, Glu, tCho and Ins for all ten volunteers. 3.3.5 Relaxation and Water Scaling Factors An example for participant-specific quantitative relaxation and water scaling maps are shown in Figure 3.6.  Figure 3.6: Sample T1 (A), T2 (B), RATTH2O (C), and [H2O] (D) maps for one subject.                                          ( A ) T1 (s) ( B ) T2 (ms)( C ) RATTH2O (%) ( D ) WC (%)%%%%%sssss%%%%%%msmsmsmsB D 79 3.3.5.1 Relaxation Correction Table 3.4 and Table 3.5 list whole brain WM and cortical GM specific relaxation measurements for each subject, respectively. The median T1 across all volunteers in WM was T1 (range) = 0.951 (0.913-0.979) s and appeared similar compared to the literature relaxation values used for the segmentation approach (T1,WM,SEG = 0.980 s) (77). A Wilcoxon signed-rank test, however, revealed that they are significantly different (p = 0.002). A similar pattern was apparent in GM median T1,GM,TWC (range) = 1.35 (1.27-1.39) s, T1,GM,SEG = 1.42 s, p = 0.002). The median WM TWC T2 (T2,WM,TWC = 92 (84-100) ms) also showed a tendency to be higher in comparison to relaxation constants used for the segmentation approach (T2 = 70.0 ms, p = 0.002) in WM as well as GM (T2,GM,TWC = 117 (105-129) ms, T2,GM,SEG = 75 ms, p = 0.002). The differences also propagate when comparing RTWC to RSEG for WM and GM. A hypothetical voxel with a WM partial volume fraction of 100 %, would demonstrate an RATTH2O,WM,SEG of 𝑅𝐴𝑇𝑇𝐻2𝑂(𝑇𝐸 = 31ms, 𝑇𝑅 = 4s, 𝑇1 = 0.980 s, 𝑇2 = 70.0ms) = 𝑒−𝑇𝐸𝑇2 ⋅ (1 − 𝑒−𝑇𝑅𝑇1) ≈ 63.1%. Similarly, a theoretical voxel located solely in GM, would demonstrate a relaxation attenuation factor of RATTH2O,GM,SEG ≈ 62.2%. In contrast, the median relaxation correction factors measured in this cohort in WM and GM were RATTH2O,WM,TWC = 84 (78-90)% and RATTH2O,GM,TWC = 88 (82-93)%, respectively. For the investigated spectroscopic VOIs with mean tissue fractions as listed in Table 3.3 the relaxation attenuation would be RATTH2O,cWM,SEG=64% and RATTH2O,pGM,SEG=63%.   80 Table 3.4: Mean correction constants for each volunteer in global white matter. Volunteers T1 (s) T2 (ms) RATTH2O (%) [H2O] (%) 1 0.958 84 78 77 2 0.913 90 83 73 3 0.936 94 87 74 4 0.976 98 88 77 5 0.946 90 84 74 6 0.946 90 82 74 7 0.979 94 85 78 8 0.979 88 80 75 9 0.935 93 86 76 10 0.957 100 90 75 Median 0.951 92 84 75 Literature 0.980 70 63 74  81 Table 3.5: Mean correction constants for each volunteer in cortical grey matter. Volunteers T1 (s) T2 (ms) RATTH2O (%) WC (%) 1 1.35 105 82 85 2 1.36 122 87 82 3 1.35 120 91 83 4 1.36 118 90 85 5 1.32 111 87 80 6 1.31 112 86 81 7 1.31 117 88 84 8 1.36 109 83 82 9 1.28 128 92 82 10 1.38 129 93 84 Median 1.35 117 88 82 Literature 1.42 75 62 82  RSEG and RTWC for the cWM and pGM VOI were significantly different (p = 0.002) with a median pairwise difference of -2.1% in cWM (range = -0.3 to 3.0%) and +3.7% in pGM (range = 1.9%-7.4%) (see Figure 3.8). The Brown–Forsythe test furthermore indicates that the variances for the water relaxation in both ROIs are significantly different between quantification methods (p(cWM) = 0.017, p(pGM) = 0.003). 82  Figure 3.7: Water relaxation attenuation for both quantification methods.  * indicate significant (p=0.002) results of the Wilcoxon signed-rank test, while † denote significant Brown–Forsythe test results. 3.3.5.2 Water Content The median water content across subjects in whole brain WM voxels was found to be WCTWC (WM) = 75% (range: 73% - 78%, p = 0.027). Water content in global cortical GM ranged from 80 % in Volunteer 5 to 85 % in Volunteers 1 and 4. Across subjects, a median GM water content of WCTWC (GM) = 82 % was found. In comparison to water content used for the segmentation method, these were found to be comparable in GM (p = 0.28). Water content for each participant are shown in Table 3.4 and Table 3.5.  Within the measured VOIs (cWM, pGM) this difference presented itself as a -3.3% lower water content in cWM and -7.4% lower WC in the pGM VOI for the TWC quantification approach (Figure 3.8). When investigating the variance of water content between the quantification methods, they were found to be significantly higher for the TWC approach in the WM VOI (p < 0.001), but not the GM VOI (p = 0.46). *,†*,†cWM pGM83  Figure 3.8: Water content correction factors for both quantification methods. * indicate significant (p=0.002) results of the Wilcoxon signed-rank test, while † denote significant Brown–Forsythe test results. 3.3.6 Metabolite Concentrations Applying the estimated SEG and calculated TWC scaling factors to the LCModel-determined metabolite concentrations in reference to water resulted in concentration values shown in Figure 3.9. There was a significant difference in median metabolite concentrations for all metabolites originating from the differences in RATTH2O and WC (cWM range = 4.2% – 12.6 %, pGM range = 5.2 %-13.0%). Metabolite concentrations calculated with the TWC method were 5.4% lower in WM and 2.4% lower in GM compared to the SEG method (p=0.002). While a systematic difference in the absolute concentrations existed, there was no significant difference in the variance between the two quantification methods for any of the metabolites (p(Brown-Forsythe) > 0.37) (Table 3.6 & Table 3.7). c  pG             ( )             A      ( ) EG   C*,†*84  Figure 3.9: Absolute metabolite concentrations as calculated with SEG (orange) and TWC (blue). Significantly different (p = 0.002) medians (Wilcoxon signed-rank test) are indicated by asterisk (*). Note that despite a significant offset between the quantification methods, no significant differences between variances were present (p > 0.37).  c  pG       NAA  (mmol  )c  pG      tCr  (mmol  )         Glu  (mmol  )                   tCho  (mmol  )             Ins  (mmol  )   C EG******** **85 Table 3.6: Absolute concentrations and percent differences between the quantification methods in the central white matter (cWM) voxel.  Wilcoxon signed-rank tests are indicated by p *, while p† refers to results from Brown-Forsythe tests.  cWM [met]SEG (range), (mmol/L) [met]TWC (range), (mmol/L) Difference, (%) p *  p †  NAA 7.7 (6.9 – 8.2) 7.3 (6.5 – 7.7) -5.4 0.002 0.37 tCr 5.4 (4.6 -5.6) 5.1 (4.5 – 5.4) -5.4 0.002 0.73 Glu 4.4 (4.1 -5.5) 4.1 (3.8 – 5.1) -5.4 0.002 0.91 tCho 1.4 (1.2 – 1.5) 1.3 (1.1 – 1.5) -5.4 0.002 0.54 Ins 3.3 (3.0 – 3.7) 3.1 (2.7 – 3.4) -5.4 0.002 0.66  Table 3.7: Absolute concentrations and percent differences between the quantification methods in the parietal grey matter (pGM) voxel. p* indicate Wilcoxon signed-rank tests are indicated by p *, while p† refers to results from Brown-Forsythe tests.   pGM [met]SEG (range), (mmol/L) [met]TWC (range), (mmol/L) Difference (%) p *  p †  NAA 9.5 (7.9-10.4) 9.1 (7.7 -10.2) -2.4 0.002 0.76 tCr 7.8 (6.8-8.7) 7.5 (6.5 – 8.5) -2.4 0.002 0.94 Glu 8.9 (7.8-10.0) 8.5 (7.6 – 9.8) -2.4 0.002 0.94 tCho 1.3 (1.1 -1.6) 1.2 (1.0 – 1.6) -2.4 0.002 0.80 Ins 5.1 (4.5-5.5) 4.9 (4.3 – 5.3) -2.4 0.002 0.66  3.4 Discussion We investigated the feasibility of mapping MRS water scaling factors on a per-subject basis for ten young healthy adults using a novel approach that corrects for subject specific water content as well as water relaxation attenuation due to water T1 and T2 and compared it to an 86 established segmentation-based approach which uses population based tissue correction factors. 3.4.1 Scan Time  The quantitative water mapping MRI protocol consisted of an eight-minute 3D GRASE sequence and a five-minute IR-T1 series, requiring 13 minutes of additional scan time beyond that used in a conventional MRS exam. In comparison, the segmentation method for water scaling only requires a high-resolution anatomical image, preferably with T1-weighting to achieve good GM/WM contrast. In our case this was achievable within seven minutes, but when working on MR systems with advanced hardware i.e. more RF receiver channels or sophisticated reconstruction algorithms allowing for sparse k-space under sampling, it should be possible to reduced further (100). While the extra time required is of patient comfort and economic concern, the additional data acquired can be used not only for calculating absolute metabolite concentrations for MRS, but also offer further valuable information about physiology or pathology. For instance, the multi-echo 3D GRASE sequence is primarily used for measuring the myelin water fraction, a marker sensitive and specific to myelin (101), which has been shown to be a useful biomarker for many diseases including multiple sclerosis, schizophrenia, phenylketonuria and autism (102).  3.4.2 Reliability of T1-Mapping The single-subject sub-study investigating the impact of flow on the T1 of CSF suggested that systematic flow within the lateral ventricles may not greatly affect T1 measurement of the CSF. This observation was based on the fact that the signal of moving water would be affected by a 87 change in gradient directions. Since scan-rescan variability appeared to be qualitatively on the same scale as variability when changing the phase encoding direction of the read-out during the IR-T1 mapping sequence, it is thus suggested that ventricular CSF flow does not greatly impact T1 measurements. These findings should however be interpreted with caution, since measurements were only taken from a single participant. Future studies including a larger sample size should compare different T1 mapping sequences e.g. variable flip angle techniques, as well as investigate the scan-rescan reproducibility of such sequences. If CSF flow remains of a concern for measuring ventricular T1, one may want to consider cardiac triggered or gated MR acquisition schemes (103).   The extended scan protocol conducted on four healthy participants demonstrated that the T1 of CSF can be estimated to be approximately 3.9 s. However, inter-subject variability was high, suggesting that it may be beneficial to prolong the shot interval of the IR-T1 mapping series used for water content mapping. Increasing the shot interval however would increase the scan time required for T1 mapping proportionally. It may prove valuable to investigate other T1 mapping methods. Of interest could be the variable flip angle technique, which could be optimized for CSF T1 measurement without the need for increasing scan time. 3.4.3 Estimate of CSF Fractions The fraction of CSF within the spectroscopic VOI was significantly lower by 12.8% when calculated with the T2 signal contribution, fCSF,S compared to the partial volume estimate based on the segmentation of T1w images, fCSF,V. A qualitative comparison of a subset of participants suggest that fCSF,S may be underestimated, since even voxels in the inter-hemispheric gap had  88 CSF contributions of less than 0.5, which should be closer to 1. This is likely due to the relatively short maximum TE sampled with the 3D GRASE acquisition of TEmax = 384ms. This is too short to accurately measure the T2 of CSF with an expected lower bound of 600ms (64). Alternatively, one could use a sequence with variable TE spacing in which a small TE spacing samples the short TE components and a larger TE spacing is used for the longer TEs (104). 3.4.4 Water Content Measurements Water content in pure segmented WM and GM resembled closely values found in the literature (64, 68). It was thus surprising to observe that water content calculated with the TWC technique varied drastically from what the segmentation technique estimated for the investigated VOIs in the cWM and pGM. It has previously been shown that T2 relaxation based TWC mapping may underestimate WC by approximately 2% (64). This could explain the lower WC in the cWM region, however, for the pGM region it is more likely that a small percentage of voxels had estimated WC larger than 100%, which can occur in voxels close or within CSF. While voxels within the lateral ventricles were eliminated when evidence of CSF flow was present in the fitting residuals, no such thresholding was conducted for CSF voxels outside the lateral ventricles, including the CSF space within the pGM region. Furthermore, the short 384 ms echo train of the GRASE sequence may not be ideal for calculating proton density nor the global geometric mean T2 of CSF, since T2 times of greater than 500 ms are to be expected. Other multi-echo T2 weighted sequences have used variable TE spacing to sample the first echo times frequently and then use longer TE intervals for later echoes (104), which may be beneficial for MRS voxels that contain a considerable CSF fractional volume component.   89 3.4.5 Absolute Metabolite Concentrations Absolute metabolite concentrations calculated with the segmentation and the TWC mapping technique had significantly different medians, as confirmed by a Wilcoxon signed-rank test.  Despite the larger variances of [H2O] and RATTH2O for the TWC approach, this increased variability did not propagate into the metabolite concentrations, which did not exhibit significantly different variances. Instead, we observed 5.4% lower metabolite concentrations in cWM and 2.4% lower in pGM for the TWC approach. This suggests that the increased variability in water scaling factors may reflect normal biological variability in healthy controls, to which the segmentation technique may not be sensitive. Furthermore, the differences observed in median metabolite concentrations could also arise from incorrectly assumed T1, T2 and WC for the specific tissue types. While the assumed tissue-specific constants may be a good estimate for a diverse sample e.g. with respect to age, the subjects participating in this study were of a narrow, young age range, which may not be reflected in the assumed correction factors used in the segmentation based approach. Both T1 and T2 have been shown to increase with age (105, 106), and would exhibit shorter relaxation values in younger adults. Incorporating age-specific relaxation values would therefore likely decrease the water relaxation attenuation for the segmentation approach and resemble the TWC approach’s water scaling factors more closely. 90 Chapter 4: Conclusion and Future Directions Phantom validations and an in vivo study have shown the future potential for measuring subject specific water relaxation and concentration maps. Subject-specific water relaxation and concentration calculations may aid in accounting for healthy biological variability and lead to a more quantitative measurement of metabolites in vivo. Differences in mean water scaling factors between the segmentation and subject-specific correction are however present. These differences arise from combination of possible inaccuracies in the assumed relaxation of tissue as well as water contents. Since the TWC method relies on an assumption of T1 in CSF for the scaling of signals in the ventricles, future work should continue the investigation of T1 measurements in the lateral ventricles and conduct further phantom validations studies measuring the metabolite signal attenuation to validate the TWC method. Once successfully implemented, the proposed technique will be especially helpful in future investigations of pathologies or anatomical locations for which reference literature values are sparse or non-existent. An example of this application can be found in a case study by Wiggerman et al investigating a severe case of radiation necrosis (107).  Furthermore, improved RF hardware including receive coils with increased FOVs and channel density covering the brain and cervical spinal cord simultaneously, permit the possible application of this technique beyond the brain, to the spinal cord as well. Only a few MRS studies have been conducted in the spinal cord, mostly reporting metabolite ratios (108–110). If GRASE and IR-T1 sequences in a sagittal orientation were developed, one could use the proposed method to calculate water content and relaxation maps for the brain and spinal cord 91 simultaneously while still using the lateral ventricles as an internal reference, thus making absolute quantification of metabolite concentrations in the spinal cord a feasible approach. The approach will also be transferable to quantification of chemical shift imaging (CSI). CSI is often used for pathologies without any or only limited applicable literature values, especially when tissue properties are varying greatly within the pathological tissue, e.g. in tumours (111). Since water content and relaxation behaviour are mapped on a voxel-by-voxel basis across the cerebrum, the TWC method could be a future application.   92 Bibliography 1.  Levitt MH: Spin Dynamics: Basics of Nuclear Magnetic Resonance. 2002. 2.  Brown RW, Cheng YCN, Haacke EM, Thompson MR, Venkatesan R: Magnetic Resonance Imaging: Physical Principles and Sequence Design: Second Edition. 2014. 3.  Bernstein MA, King KF, Zhou XJ: Handbook of MRI Pulse Sequences. 2004. 4.  de Graaf RA: In Vivo NMR Spectroscopy: Principles and Techniques. 2nd edition. Chichester, West Sussex, United Kingdom: John Wiley & Sons, Ltd; 2007. 5.  de Graaf RA: In Vivo NMR Spectroscopy - Static Aspects (Proton NMR Spectroscopy). In Vivo NMR Spectrosc Princ Tech. 2nd edition. Chichester, West Sussex, United Kingdom: John Wiley & Sons, Ltd; 2007:43–78. 6.  Spector R, Robert Snodgrass S, Johanson CE: A balanced view of the cerebrospinal fluid composition and functions: Focus on adult humans. Exp Neurol 2015; 273:57–68. 7.  Lisanti C, Carlin C, Banks KP, Wang D: Normal MRI Appearance and Motion-Related Phenomena of CSF. Am J Roentgenol 2007; 188:716–725. 8.  Servier Medical ART (SMART) [https://smart.servier.com/] 9.  Myelinated neuron [https://commons.wikimedia.org/wiki/File:Myelinated_neuron.jpg] 10.  Edden RAE, Harris AD: The Significance of N-Acetyl Aspartate in Human Brain MRS. In Handb Magn Reson Spectrosc Vivo MRS Theory, Pract Appl. 1st edition. Edited by Bottomley PA, Griffiths JR. Chichester, West Sussex, United Kingdom: John Wiley & Sons; 2016:947–956. 11.  Ramadan S, Lin A, Stanwell P: Glutamate and glutamine: a review of in vivo MRS in the 93 human brain. NMR Biomed 2013; 26:1630–46. 12.  Rackayova V, Cudalbu C, Pouwels PJW, Braissant O: Creatine in the central nervous system: From magnetic resonance spectroscopy to creatine deficiencies. Anal Biochem 2017; 529:144–157. 13.  Öz G, Alger JR, Barker PB, et al.: Clinical proton MR spectroscopy in central nervous system disorders. Radiology 2014; 270:658–679. 14.  Emsley J: Nature’s Building Blocks: An A-Z Guide to the Elements. New Ed. New York City (NY): Oxford University Press Inc.; 2011. 15.  Bloch F: Nuclear Induction. Phys Rev 1946; 70:460–474. 16.  Hajnal J V., de Coene B, Lewis PD, et al.: High signal regions in normal white matter shown by heavily t2-weighted csf nulled ir sequences. J Comput Assist Tomogr 1992. 17.  Mugler JP, Brookeman JR: Rapid three-dimensional T1-weighted MR imaging with the MP-RAGE sequence. J Magn Reson Imaging 1991; 1:561–567. 18.  MacKay A, Whittall K, Adler J, Li D, Paty D, Graeb D: In vivo visualization of myelin water in brain by magnetic resonance. Magn Reson Med 1994; 31:673–677. 19.  Whittall KP, MacKay AL, Graeb DA, Nugent RA, Li DKB, Paty DW: In vivo measurement of T2 distributions and water contents in normal human brain. Magn Reson Med 1997; 37:34–43. 20.  Laule C, Kozlowski P, Leung E, Li DKB, MacKay AL, Moore GRW: Myelin water imaging of multiple sclerosis at 7 T: Correlations with histopathology. Neuroimage 2008; 40:1575–1580. 21.  Hahn EL: Spin echoes. Phys Rev 1950. 94 22.  Christensen KA, Grant DM, Schulman EM, Walling C: Optimal Determination of Relaxation Times of Fourier Transform Nuclear Magnetic Resonance. Determination of Spin-Lattice Relaxation Times in Chemically Polarized Species. J Phys Chem 1974; 78:1971–1977. 23.  Homer J, Beevers MS: Driven-equilibrium single-pulse observation of T1 relaxation. A reevaluation of a rapid “new” method for determining NMR spin-lattice relaxation times. J Magn Reson 1985; 63:287–297. 24.  P. McIntosh L: CPMG. In Encycl Biophys. Edited by Roberts GCK. Berlin, Heidelberg: Springer Berlin Heidelberg; 2013:386. 25.  Meiboom S, Gill D: Modified spin-echo method for measuring nuclear relaxation times. Rev Sci Instrum 1958; 29:688–691. 26.  Burstein D: Stimulated echoes: Description, applications, practical hints. Concepts Magn Reson 1996; 8:269–278. 27.  Prasloski T, Rauscher A, MacKay AL, et al.: Rapid whole cerebrum myelin water imaging using a 3D GRASE sequence. Neuroimage 2012; 63:533–539. 28.  Chavhan GB, Babyn PS, Thomas B, Shroff MM, Haacke EM: Principles, Techniques, and Applications of T2*-based MR Imaging and Its Special Applications. RadioGraphics 2009; 29:1433–1449. 29.  Deoni SCL, Rutt BK, Arun T, Pierpaoli C, Jones DK: Gleaning multicomponent T 1 and T 2 information from steady-state imaging data. Magn Reson Med 2008. 30.  Nguyen TD, Wisnieff C, Cooper MA, et al.: T2 prep three-dimensional spiral imaging with efficient whole brain coverage for myelin water quantification at 1.5 tesla. Magn Reson 95 Med 2012; 67:614–621. 31.  De León-Rodríguez LM, Martins AF, Pinho MC, Rofsky NM, Sherry AD: Basic MR relaxation mechanisms and contrast agent design. J Magn Reson Imaging 2015; 42:545–565. 32.  Köylü MZ, Asubay S, Yilmaz A: Determination of Proton Relaxivities of Mn(II), Cu(II) and Cr(III) added to Solutions of Serum Proteins. Molecules 2009; 14:1537–1545. 33.  Caravan P, Farrar CT, Frullano L, Uppal R: Influence of molecular parameters and increasing magnetic field strength on relaxivity of gadolinium- and manganese-based T 1 contrast agents. Contrast Media Mol Imaging 2009; 4:89–100. 34.  Govindaraju V, Young K, Maudsley AA: Proton NMR chemical shifts and coupling constants for brain metabolites. NMR Biomed 2000; 13:129–153. 35.  Harris AD, Saleh MG, Edden RAE: Edited 1H magnetic resonance spectroscopy in vivo: Methods and metabolites. Magn Reson Med 2017; 77:1377–1389. 36.  Bottomley P: Selective volume method for performing localized NMR spectroscopy. United States Pat Trademark Off 1985:iv--v. 37.  Frahm J, Merboldt K-D, Hänicke W: Localized proton spectroscopy using stimulated echoes. J Magn Reson 1987; 72:502–508. 38.  Ordidge RJ, Connelly A, Lohman JAB: Image-selected in Vivo spectroscopy (ISIS). A new technique for spatially selective nmr spectroscopy. J Magn Reson 1986; 66:283–294. 39.  Slotboom J, Mehlkopf A., Bovée WMM.: A single-shot localization pulse sequence suited for coils with inhomogeneous RF fields using adiabatic slice-selective RF pulses. J Magn Reson 1991; 95:396–404. 96 40.  Sacolick LI, Rothman DL, de Graaf RA: Adiabatic refocusing pulses for volume selection in magnetic resonance spectroscopic imaging. Magn Reson Med 2007; 57:548–553. 41.  Klose U: In vivo proton spectroscopy in presence of eddy currents. Magn Reson Med 1990; 14:26–30. 42.  Provencher SW: Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn Reson Med 1993; 30:672–679. 43.  Wilson M, Reynolds G, Kauppinen RA, Arvanitis TN, Peet AC: A constrained least-squares approach to the automated quantitation of in vivo 1 H magnetic resonance spectroscopy data. Magn Reson Med 2011; 65:1–12. 44.  Zoelch N, Hock A, Henning A: Quantitative magnetic resonance spectroscopy at 3T based on the principle of reciprocity. NMR Biomed 2018; 31:e3875. 45.  Barker PB, Soher BJ, Blackband SJ, Chatham JC, Mathews VP, Bryan RN: Quantitation of Proton NMR-Spectra of the Human Brain Using Tissue Water as an Internal Concentration Reference. NMR Biomed 1993; 6:89–94. 46.  Alger JR: Quantitative proton magnetic resonance spectroscopy and spectroscopic imaging of the brain: A didactic review. Top Magn Reson Imaging 2010; 21:115–128. 47.  de Graaf RA: In Vivo NMR Spectroscopy: Principles and Techniques. 3rd edition. Chichester, West Sussex, United Kingdom: John Wiley & Sons Ltd; 2019. 48.  Cady EB: External Concentration References. In Clin Magn Reson Spectrosc. New York City (NY): Plenum Press; 1990:277. 49.  Barantin L, Le Pape A, Akoka S: A new method for absolute quantitation of MRS metabolites. Magn Reson Med 1997; 38:179–182. 97 50.  Tofts PS, Waldman AD: Spectroscopy: 1H Metabolite Concentrations. In Quant MRI Brain. Edited by Tofts P. Chichester, UK: John Wiley & Sons, Ltd; 2004:299–339. 51.  Cecil KM, Hills EC, Sandel ME, et al.: Proton magnetic resonance spectroscopy for detection of axonal injury in the splenium of the corpus callosum of brain-injured patients. J Neurosurg 1998; 88:795–801. 52.  Sajja BR, Wolinsky JS, Narayana PA: Proton Magnetic Resonance Spectroscopy in Multiple Sclerosis. Neuroimaging Clin N Am 2009; 19:45–58. 53.  Hoch SE, Kirov II, Tal A: When are metabolic ratios superior to absolute quantification? A statistical analysis. NMR Biomed 2017; 30:e3710. 54.  MacMillan EL, Schuber JJ, Vavasour IM, et al.: N-acetylaspartate and creatine decrease with myelin damage in relapsing multiple sclerosis. In Proc Congr Eur Comm Treat Res Mult Scler. Barcelona: ECTRIMS Online Library; 2015:P422. 55.  Maudsley AA, Domenig C, Govind V, et al.: Mapping of brain metabolite distributions by volumetric proton MR spectroscopic imaging (MRSI). Magn Reson Med 2009; 61:548–559. 56.  LCModel & LCMgui User’s Manual [http://s-provencher.com/pub/LCModel/manual/manual.pdf] 57.  Ernst T, Kreis R, Ross BD: Absolute Quantitation of Water and Metabolites in the Human Brain. I. Compartments and Water. J Magn Reson Ser B 1993; 102:1–8. 58.  Laule C, Vavasour IM, Moore GRW, et al.: Water content and myelin water fraction in multiple sclerosis. A T2 relaxation study. J Neurol 2004; 251:284–93. 59.  Lee D-H, Heo H-Y, Zhang K, et al.: Quantitative assessment of the effects of water proton 98 concentration and water T 1 changes on amide proton transfer (APT) and nuclear overhauser enhancement (NOE) MRI: The origin of the APT imaging signal in brain tumor. Magn Reson Med 2017; 77:855–863. 60.  Öz G: Magnetic Resonance Spectroscopy of Degenerative Brain Diseases. 2016. 61.  Papadopoulos MC, Saadoun S, Binder DK, Manley GT, Krishna S, Verkman AS: Molecular mechanisms of brain tumor edema. Neuroscience 2004:1011–1020. 62.  Oeltzschner G, Schnitzler A, Wickrath F, Zöllner HJ, Wittsack HJ: Use of quantitative brain water imaging as concentration reference for J-edited MR spectroscopy of GABA. Magn Reson Imaging 2016; 34:1057–1063. 63.  Sabati M, Govind V, Maudsley AA: Signal Normalization for MR Spectroscopic Imaging Using Brain Tissue Water: Variability and Pathologic Detectability. In Proc 19th Annu Meet ISMRM, Montr Canada. Montreal; 2011:1434. 64.  Meyers SM, Kolind SH, MacKay AL: Simultaneous measurement of total water content and myelin water fraction in brain at 3T using a T2 relaxation based method. Magn Reson Imaging 2017; 37:187–194. 65.  Meyers SM, Kolind SH, Laule C, MacKay AL: Measuring water content using T2 relaxation at 3T: Phantom validations and simulations. Magn Reson Imaging 2016; 34:246–251. 66.  Meyers SM: Accurate measurement of brain water content by magnetic resonance. University of British Columbia; 2015(August). 67.  Zhang J, Vavasour IM, Kolind SH, Baumeister R, Rauscher A, MacKay AL: Advanced Myelin Water Imaging Techniques for Rapid Data Acquisition and Long T2 Component Measurements. In Proc 23rd Annu Meet ISMRM, Toronto, Canada. Milano; 2015:824. 99 68.  Neeb H, Zilles K, Shah NJ: A new method for fast quantitative mapping of absolute water content in vivo. Neuroimage 2006; 31:1156–1168. 69.  Maedler B, Trudy H, MacKay AL: 3D-relaxometry - Quantitative T1 and T2 brain mapping at 3T. In Proc 14th Annu Meet ISMRM, Seattle, USA. Seattle; 2006:958. 70.  Whittall KP, MacKay AL: Quantitative interpretation of NMR relaxation data. J Magn Reson 1989; 84:134–152. 71.  Prasloski T, Mädler B, Xiang QS, MacKay A, Jones C: Applications of stimulated echo correction to multicomponent T2 analysis. Magn Reson Med 2012; 67:1803–1814. 72.  Mikkelsen M, Barker PB, Bhattacharyya PK, et al.: Big GABA: Edited MR spectroscopy at 24 research sites. Neuroimage 2017; 159:32–45. 73.  Lebel RM, Wilman AH: Transverse relaxometry with stimulated echo compensation. Magn Reson Med 2010; 64:1005–1014. 74.  Smith SA, Levante TO, Meier BH, Ernst RR: Computer Simulations in Magnetic Resonance. An Object-Oriented Programming Approach. J Magn Reson Ser A 1994; 106:75–105. 75.  McCarthy P: FSLeyes. 2018. 76.  Fedorov A, Beichel R, Kalpathy-Cramer J, et al.: 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging 2012; 30:1323–1341. 77.  Vavasour IM, Meyers SM, Mädler B, et al.: Multicenter Measurements of T 1 Relaxation and Diffusion Tensor Imaging: Intra and Intersite Reproducibility. J Neuroimaging 2019; 29:42–51. 78.  Meyers SM, Laule C, Vavasour IM, et al.: Reproducibility of myelin water fraction analysis: a comparison of region of interest and voxel-based analysis methods. Magn 100 Reson Imaging 2009; 27:1096–1103. 79.  Meyers SM, Vavasour IM, Madler B, et al.: Multicenter measurements of myelin water fraction and geometric mean T2: Intra- and intersite reproducibility. J Magn Reson Imaging 2013; 38:1445–1453. 80.  Mitchell MD, Kundel HL, Axel L, Joseph PM: Agarose as a tissue equivalent phantom material for NMR imaging. Magn Reson Imaging 1986; 4:263–266. 81.  Hellerbach A, Schuster V, Jansen A, Sommer J: MRI Phantoms - Are There Alternatives to Agar? PLoS One 2013; 8:e70343. 82.  Portakal ZG, Shermer S, Jenkins C, et al.: Design and characterization of tissue-mimicking gel phantoms for diffusion kurtosis imaging. Med Phys 2018; 45:2476–2485. 83.  Dickinson RJ, Hall AS, Hind AJ, Young IR: Measurement of changes in tissue temperature using MR imaging. J Comput Assist Tomogr 1986; 10:468–72. 84.  Deoni SCL: High-resolution T1 mapping of the brain at 3T with driven equilibrium single pulse observation of T1 with high-speed incorporation of RF field inhomogeneities (DESPOT1-HIFI). J Magn Reson Imaging 2007; 26:1106–1111. 85.  Madsen KS, Holm DA, Søgaard LV, Rowland IJ: Effect of paramagnetic manganese cations on 1H MRS of the brain. NMR Biomed 2008; 21:1087–1093. 86.  Lenkinski RE, Wang X, Elian M, Goldberg SN: Interaction of gadolinium-based MR contrast agents with choline: Implications for MR spectroscopy (MRS) of the breast. Magn Reson Med 2009; 61:1286–1292. 87.  Jansen JF a, Backes WH, Nicolay K, Kooi ME: 1H MR spectroscopy of the brain: absolute quantification of metabolites. Radiology 2006; 240:318–332. 101 88.  Kreis R, Ernst T, Ross BD: Absolute Quantitation of Water and Metabolites in the Human Brain. II. Metabolite Concentrations. J Magn Reson Ser B 1993; 102:9–19. 89.  Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW, Smith SM: FSL. Neuroimage 2012; 62:782–790. 90.  Smith SM: Fast robust automated brain extraction. Hum Brain Mapp 2002; 17:143–155. 91.  Zhang Y, Brady M, Smith S: Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans Med Imaging 2001; 20:45–57. 92.  Kempton MJ, Underwood TSA, Brunton S, et al.: A comprehensive testing protocol for MRI neuroanatomical segmentation techniques: Evaluation of a novel lateral ventricle segmentation method. Neuroimage 2011; 58:1051–1059. 93.  The FIL Methods Group: SPM8 Manual. 2013:475. 94.  Jost G, Harting I, Heiland S: Quantitative single-voxel spectroscopy: The reciprocity principle for receive-only head coils. J Magn Reson Imaging 2005; 21:66–71. 95.  Kreis R: The trouble with quality filtering based on relative Cramér-Rao lower bounds. Magn Reson Med 2016; 75:15–18. 96.  Wilcoxon F: Individual Comparisons by Ranking Methods. Biometrics Bull 1945; 1:80. 97.  Brown MB, Forsythe AB: Robust tests for the equality of variances. J Am Stat Assoc 1974; 69:364–367. 98.  Spearman C: The Proof and Measurement of Association between Two Things. Am J Psychol 1904; 15:72–101. 99.  Lin C, Bernstein M, Huston J, Fain S: Measurements of T1 Relaxation times at 3.0T: 102 Implications for clinical MRA. In Proc 9th Annu Meet ISMRM, Glas Scotl. Glasgow; 2001:1391. 100.  Geethanath S, Reddy R, Konar AS, et al.: Compressed sensing MRI: a review. Crit Rev Biomed Eng 2013; 41:183–204. 101.  Laule C, Yung A, Pavolva V, et al.: High-resolution myelin water imaging in post-mortem multiple sclerosis spinal cord: A case report. Mult Scler 2016; 22:1485–1489. 102.  MacKay AL, Laule C: Magnetic Resonance of Myelin Water: An in vivo Marker for Myelin. Brain Plast 2016; 2:71–91. 103.  Suzuka T, Nagai H, Ohara S, Banno T: Observation of the CSF Pulsatile Flow on MRI (1): ECG-triggered MRI and CSF pulsatile flow. In Annu Rev Hydroceph. Berlin, Heidelberg: Springer Berlin Heidelberg; 1990:53–54. 104.  Laule C, Vavasour IM, Kolind SH, et al.: Long T2 water in multiple sclerosis: what else can we learn from multi-echo T2 relaxation? J Neurol 2007; 254:1579–1587. 105.  Callaghan MF, Freund P, Draganski B, et al.: Widespread age-related differences in the human brain microstructure revealed by quantitative magnetic resonance imaging. Neurobiol Aging 2014; 35:1862–1872. 106.  Kumar R, Delshad S, Woo MA, MacEy PM, Harper RM: Age-related regional brain T2-relaxation changes in healthy adults. J Magn Reson Imaging 2012; 35:300–308. 107.  Wiggermann V, Lapointe E, Litvin L, et al.: Longitudinal advanced <scp>MRI</scp> case report of white matter radiation necrosis. Ann Clin Transl Neurol 2018:acn3.704. 108.  Hock A, Henning A, Boesiger P, Kollias SS: (1)H-MR spectroscopy in the human spinal cord. AJNR Am J Neuroradiol 2013; 34:1682–1689. 103 109.  Hock A, MacMillan EL, Fuchs A, et al.: Non-water-suppressed proton MR spectroscopy improves spectral quality in the human spinal cord. Magn Reson Med 2013; 69:1253–1260. 110.  Holly LT, Freitas B, McArthur DDL, Salamon N: Proton magnetic resonance spectroscopy to evaluate spinal cord axonal injury in cervical spondylotic myelopathy: Laboratory investigation. J Neurosurg Spine 2009; 10:194–200. 111.  Kurhanewicz J, Vigneron DB, Nelson SJ: Three-Dimensional Magnetic Resonance Spectroscopic Imaging of Brain and Prostate Cancer. Neoplasia 2000; 2:166.  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0380533/manifest

Comment

Related Items