Search

Small vessel disease (SVD) is a common cause of strokes and dementia. Currently, there are no treatments; therefore, developing and validating early biomarkers of disease progression and treatment response is important for future drug trials. Though SVD pathogenesis is not well understood, findings from previous studies suggest that blood-brain barrier dysfunction and impaired cerebrovascular reactivity (CVR) contribute to the disease. The latter can be measured in vivo using a vasoactive stimulus in parallel with magnetic resonance imaging (MRI) techniques sensitive to blood flow, such as blood oxygen level dependent (BOLD) contrast, and has frequently been assessed in patients with steno-occlusive diseases. However, it is unclear if the technique is reliable when investigating cerebrovascular health in deep structures of the brain where SVD is prevalent. Therefore, this thesis aimed to assess and optimise the reliability of CVR measurements and deepen our understanding of its role in SVD pathogenesis. A systematic review was performed to provide a detailed overview of CVR MRI methodologies and clinical applications, including SVD, present in the literature, which identified several acquisition and analysis methods, a need for greater standardisation and lack of data on reliability. Specifically in SVD research, there was limited application of CVR MRI in SVD populations, little optimisation and reliability assessment of CVR in deep brain structures relevant to SVD, such as in white and subcortical grey matter. Following those findings, the effects of voxel- and region-based analysis approaches on reliability of CVR estimates were investigated using simulations and test-retest data from healthy volunteers. Voxel-based CVR magnitude estimates in tissues with high noise levels were prone to bias, whereas biases in region-based estimates were independent of noise level, but consistently underestimated CVR magnitude relative to the ground-truth mean. Furthermore, the test-retest study confirmed the repeatability of CVR estimates from a BOLD-CVR experiment with fixed inhaled stimulus, although a systematic, but small, bias was detected due to habituation to the gas challenge. The data from healthy volunteers were further used to conduct a proof-of-concept and investigate the feasibility of extracting cerebral pulsatility from BOLD-CVR data. Small-to-moderate correlations with pulsatility from phase-contrast MRI were found depending on the regions considered. CVR pulsatility was also computed in a small cohort of SVD patients: it was higher than in healthy volunteers, but no associations were found with SVD burden. It was concluded that further optimisation and validation of the technique is needed before being suitable for clinical research. Following the optimisation of the CVR MRI technique, relationships between CVR and SVD neuroimaging features, cognition, stroke severity and outcome were investigated cross-sectionally and longitudinally in a cohort of patients with mild stroke. In the cross-sectional analysis, CVR impairment in normal-appearing and damaged tissues was associated with worse SVD burden and cognition deficit. Furthermore, the longitudinal analysis showed that baseline CVR impairment predicted worsening of white matter hyperintensity and perivascular space volumes after one year. In conclusion, assessment of CVR in the brain and its deeper structures was successfully conducted in healthy volunteers and patients with SVD using MRI. However, this required appropriate optimisation of processing strategy as the latter can affect accuracy of CVR parameters and inter-study comparability. Importantly, applying the technique in a cohort of SVD patients led to the findings that CVR impairment was related to worse SVD burden and is a potential marker of SVD severity and progression.

Introduction: Non-ischaemic dilated cardiomyopathy (DCM) is defined as dilatation and systolic impairment of the left ventricle in the absence of significant epicardial coronary artery disease or abnormal loading conditions. DCM is common condition with a prevalence of 1 in 250 and has substantial mortality through sudden cardiac death and pump failure. Current risk prediction models are suboptimal, and further work is required to find predictors of adverse outcome. Global longitudinal strain (GLS) has had conflicting results as to whether it determines prognosis in DCM and further work is needed to clarify this. Myocardial blood flow (MBF) has been found to be abnormal in DCM in a previous study using cardiovascular magnetic resonance (CMR), but has not been established with the novel Kellman quantitative perfusion technique. Furthermore, the determinants of impaired MBF in DCM are not well described. Although exercise capacity is reduced in DCM, the determinants of this are unclear. MBF has been found to be an important predictor of exercise capacity in other cardiac conditions. Methods: Firstly, a systematic review was undertaken of several established databases of medical literature to determine whether GLS was predictive of prognosis in DCM. Subsequently, 37 patients with DCM were prospectively recruited into the main study. These patients underwent cardiopulmonary exercise testing and comprehensive CMR imaging with GLS assessment and quantitative myocardial perfusion. Results: Our systematic review yielded 8 studies: 4 studies had an association between GLS and clinical outcomes in DCM and 4 studies did not. The larger higher quality studies tended to favour an association between abnormal GLS and clinical outcomes. In the prospective study, DCM patients had reduced myocardial perfusion reserve (MPR) compared to healthy controls (1.97±0.65 vs 3.08±0.79, respectively; p < 0.001). MPR was a significantly correlated with % predicted peak VO2 (r=0.385; p=0.019). This remained significant when assessed in all multivariable models. MPR did not correlate with CMR markers of fibrosis. Conclusions: The systematic review provides evidence to support GLS as a predictor of prognosis in DCM. However further larger prospective studies are required. MPR is reduced in patients with DCM when compared to controls. MPR is a significant predictor of percentage predicted peak VO2 in patients with DCM. CMR markers of fibrosis do not appear to contribute to reduced MPR in patients with DCM.

Back muscle functionality is important in maintaining stability and balance. This thesis assesses the quality of data that can be obtained from various methods used to examine back muscle function. Magnetic resonance imaging (MRI) can be used to obtain high quality structural images of spinal muscles and from these, cross-sectional areas and volumes can be determined, which are frequently utilised as a direct indicator of muscle strength and functionality. Whether the values obtained varied with the scanning position, and specifically degree of spinal curvature, was examined to determine what impact this might have when undertaking longitudinal measurements. Overall, it was found that spinal muscle volume determinations are robust in terms of their sensitivity to participant position and spinal curvature and these factors should not have a major impact on the results in intervention studies. Magnetic resonance spectroscopy (MRS) was used to determine the phosphocreatine (PCr) recovery rate constant in the spinal muscles following exercise, as a measure of oxidative capacity. High levels of reliability were found for the parameter when measured using a 3T scanner indicating its suitability for monitoring changes in spinal muscle functionality when undertaking longitudinal measurements. Near infrared spectroscopy (NIRS) was used to determine the deoxyhaemoglobin recovery time constant following exercise. NIRS is a technique which is sensitive to tissue location, and it was found that in approximately 30% of the data sets acquired the signal was of too low quality to undertake fitting of the recovery data. This was most likely because of relatively high levels of subcutaneous fat being present in the back. In those cases where data fitting was possible, however, high reliability was found. Due to Covid-19 restrictions it was not possible to undertake an intervention study using the above techniques to assess their ability to detect muscle functionality changes as originally planned. Therefore, it was decided to investigate whether there were alternative indirect measurements that could potentially be used to assess spinal muscle function. To this end, a pre-existing data set was used to examine the relationship between hand-grip strength (HGS) measurements and spinal muscle functionality, with the latter assessed via cross-sectional area measurements. Although a significant correlation was found between the two, it was determined that HGS was no better a predictor of spinal muscle size than simple anthropometry measures such as age, height, and weight. Such a finding supports the need for more direct measures of spinal muscle function if this is to be accurately monitored.

Metabolic diseases have reached pandemic levels, raising concerns about metabolic health. Their prevalence is expected to rise in the future. Ramadan fasting (RF), a yearly Islamic practice, involves abstaining from eating and drinking from sunrise until sunset. The effects of RF vary due to multiple factors, including the duration of fasting per day, variations in diet quality and quantity, and the level of physical activity observed during Ramadan, leading to heterogeneity in reported outcomes. The main aim of the thesis was to test the hypothesis that 18 hrs/day RF causes systemic metabolic changes, which could be detected non-invasively by MRI and basic blood samples. The secondary aims were to study these changes on healthy female and male groups. Als, as part of this assessment, the validation of automated muscle and fat volumes measurements method and the optimisation of MR image acquisitions had to be undertaken to improve the automated measurements. Methods: The validation of automated muscle volume measurements involved comparing them with the gold standard manual method in a population undergoing temporal changes in muscle volume. Subsequently, various MR sequences were assessed to identify the ones that produced images with the highest contrast-to-noise ratio and were most compatible with the automated threshold method. To evaluate the effect of RF on the body mass index (BMI), muscle and subcutaneous adipose tissue volumes, liver and muscle ectopic lipids, abdominal visceral and subcutaneous fat (AVF and ASF), fasting blood glucose and insulin (FBG and FBI), and blood lipid profiles were monitored before, during, and a month after Ramadan. Additionally, the dietary intake and physical activity habits were assessed during and one month after the end of RF. Results: The automated and manual measurements of muscle volume exhibited strong agreement. Optimal results were obtained using fat-only and water-only images for automated measurements of muscle and fat volumes. After RF, significant reductions were observed in BMI and skeletal muscle volume, while FBG, LDL, and ectopic muscle lipid increased in both males and females. AVF significantly decreased in males, and there was a tendency towards a decrease in ASF in both sexes. Although protein intake did not fall below the recommended guideline level, it was significantly lower during RF compared to the non-RF period. Discussion: The study findings indicate that automated volume measurements demonstrate comparability to the gold standard method while saving time. The declines in BMI and abdominal adipose fats are factors in improving cardiometabolic health. The adverse increases in FBG, LDL, and ectopic muscle lipid could be attributed to circadian rhythm disruption, which impacts hormonal levels, lipolysis and hepatic gluconeogenesis. The decrease in muscle volume may be associated with the limited eating window during fasting. Conclusion: An 18-hour RF regimen can lead to reductions in BMI and adipose fat tissues, although caution is warranted regarding the effects on muscle volume, FBG, and LDL. To counteract these effects, aligning with the wider literature, it is recommended to enhance protein intake and include resistance exercise. Furthermore, the study underscores the reliability of automated volume measurements as a time-efficient substitute for manual methods.

Cerebral small vessel disease (SVD) is the breakdown of the small blood vessels of the brain, leading to many cases of stroke and dementia. The pathophysiology of SVD is largely unknown, although several mechanisms have been suggested. One such mechanism is the role of increased blood flow pulsatility into the brain, caused by vessel stiffening, leading to damage of the microvasculature. Magnetic resonance imaging (MRI) allows us to non-invasively measure blood flow and velocity using a technique called phase contrast-MRI - traditionally used with 2D slices across the vessel(s) of interest. An advanced form of phase-contrast MRI, known as 4D flow, has emerged in recent years that allows for a volume of data to be acquired, containing velocity information in all directions. However, to keep scan times practical when collecting this amount of data, spatiotemporal resolution has to be sacrificed. The main aim of this thesis was to assess 4D flow's capabilities, including comparing it to the more well-established 2D method in healthy volunteers, patients, and phantom experiments, so as to better understand its role in investigating SVD. Another aim was to learn more about the role of flow and pulsatility in SVD development in patients using data acquired in the longitudinal Mild Stroke Study 3 (MSS3). Firstly, I systematically reviewed studies that have assessed the human brain using 4D flow. Across 61 relevant studies, I found a general consensus for the current use of the technique in this context. I then optimised the Siemens prototype 4D flow sequence (N = 11 healthy volunteers), testing different parameters to find the combination that best balanced scan quality and duration. I then assessed the test-retest repeatability and intra-rater reliability of both 2D and 4D methods (N = 11 healthy volunteers), as well as differences between them. Following this, I performed the same 4D-2D comparison on SVD patients (N = 10). Absolute flow measurements using 4D flow were shown to have moderate repeatability and reliability, while flow pulsatility measurements showed acceptable repeatability and reliability. Furthermore, 2D arterial pulsatility was measured higher than with 4D, while 4D often measured higher flow rates than 2D. 4D flow was shown to be feasible when used on SVD patients, with no noticeable issues caused by potential patient movement. Flow data analysis from the longitudinal SVD study MSS3 showed that intracranial pulsatility is associated with cross-sectional SVD lesion volume but not longitudinal lesion growth, with stronger associations seen in the arteries of the neck compared to the venous sinuses.

Prader-Willi syndrome (PWS) is a neurodevelopmental disorder resulting from the absent expression of maternally imprinted, paternally expressed genes located in the chromosomal region 15q11-13. This absence of expression is usually either due to a paternal deletion or maternal uniparental disomy (mUPD). Both genetic subtypes share the same core clinical symptoms, such as developmental delays, intellectual disability, hyperphagia, behavioural and social difficulties, and a heightened risk for anxiety and depression. However, only people with PWS due to mUPD seem to be at particularly high risk for developing psychosis. Research on the causes of this differential risk is very scarce. This PhD study aims to propose potential mechanisms of psychosis in PWS and design and develop a study to test them. I hypothesised that the GABA/glutamate equilibrium is disrupted in PWS due to the absence of paternal expression. In addition, the overexpression of maternal genes on chromosome 15 would cause cholinergic dysfunction in people with mUPD and further disrupt GABAergic, glutamatergic, dopaminergic and cholinergic functions, increasing the probability of psychotic symptoms. In line with the predictive processing model of psychosis, the neurotransmitter disruptions proposed would cause a reduction in the use of priors to formulate predictions, detect and generate prediction errors, and update the prediction system. A case-control study comprising clinical, cognitive, neuroimaging, and remote behavioural testing was designed to investigate the hypotheses. The study was organised into four modules. The EEG module comprised the Global-Local paradigm, a sensory gating paradigm, and resting state. The MRS module measured metabolites in the anterior cingulate cortex, the auditory and visual cortex. The cognitive module contained measures of IQ, processing speed and working memory. The behavioural module was added during the COVID-19 pandemic to enable online data collection and focused on measuring the use of priors in perception. All modules were assessed for use in a PWS population and were feasible, with adjustments. Nineteen participants took part in at least part of the study, demonstrating its feasibility in a PWS population. Because of time constraints, COVID-19, and resources available, not all participants completed all measures. For the same reasons, only the EEG Global-Local data was analysed. The EEG paradigms were collected on 19 participants (11 with deletion, 4 with mUPD), and the global-local paradigm was analysed. Preliminary analysis revealed that people with mUPD have smaller P300 in response to an oddball paradigm than deletion and control participants. Participants with deletion have smaller P300 responses than controls showing impaired attention switching and a flaw in the predicting coding system of people with PWS, particularly in those with mUPD. Contrary to what was expected, the mismatch negativity response to the oddball paradigm was not different between mUPD and controls but was significantly earlier in the deletion group. This could indicate a chronic fight or flight state having the potential to disrupt neurotransmission, interoception and perception further, causing additional disruption of the predictive processing system. A planned analysis of the remaining measures is presented in this thesis and will be applied in the future when sufficient data is analysed.

Adolescence is a period of significant neurodevelopment and increased vulnerability to the onset of depression. However, the neural underpinnings of depression during adolescence and the associated risk factors are not well understood. The aim of this PhD research was to fill this knowledge gap by examining biological and psychosocial factors associated with the emergence of depression during adolescence. Using a large, population-based sample, the Adolescent Brain Cognitive Development (ABCD) Study, my doctoral work found that depression in early adolescence is associated with similar neuroimaging findings (cortical and white matter microstructural features) to those seen in adult depression samples. Further, the work in this thesis demonstrated that earlier pubertal timing is associated with an increased risk for later depression in adolescence. While earlier pubertal timing was also related to structural brain features, brain structure was not found to mediate the observed association between early pubertal timing and later depressive symptoms. This finding highlights the important role that other aspects of a young person's biology, psychology and social world may play, and should be explored in future work. This thesis also investigated how dynamic functional brain networks relate to irritability in adolescent depression using a co-produced youth-researcher design. In this pilot study, I first worked with young people to develop a novel fMRI irritability task that reflected the social nature of irritability in adolescence. Using a local sample of youth with depressive symptoms, I found that dynamic functional brain networks differed between the irritability task and a standard resting state scan, which provides preliminary evidence for validation of this novel task. Finally, my work demonstrated that properties of dynamic brain networks related to emotion regulation and cognitive control were associated with youth depressive symptoms and irritable mood. Taken together, the findings of this thesis suggest that neuroanatomical differences may be present early in the disease course of depression and that biological factors, such as early pubertal development, relate to depression risk. Moreover, this work provides preliminary evidence to suggest that alterations in dynamic brain network properties are associated with depressive symptoms and irritability in adolescence. Further, this doctoral research highlights the importance of co-produced study designs in developmental cognitive neuroscience. This work makes an important contribution to our understanding of the factors associated with the emergence of depression during adolescence, which lays a strong foundation upon which to base future longitudinal research.

In this thesis a range of catalyst pellets are studied using Nuclear Magnetic Resonance (NMR) and Magnetic Resonance Imaging (MRI) techniques with the aim of characterizing mass transport in the pellets and exploring the presence of spatial heterogeneities formed in the manufacturing process. Measurements have been done both in bulk, i.e. spatially unresolved, and at high spatial resolution. Measurements have been used for comparison with standard bulk pellet characterization methods, in particular the estimation of the tortuosity via the bulk porosity, but also to provide information about whether these characteristics are uniform throughout the pellet or whether there are heterogeneities introduced by the manufacturing process. Using the insights gained into mass transport characteristics and spatial heterogeneities in catalyst pellets should lead to better understanding and optimisation of catalysts choice and catalyst production processes (mainly extrusion, drying and calcination) and models of catalytic systems. Spatially unresolved Pulsed Field Gradient NMR (PFG NMR) methods have been used to measure the self diffusion coefficient of liquid contained within catalyst support pellets. Mass transport in catalyst pellets is often characterized through the tortuosity parameter, which impacts the catalyst effectiveness, though it is difficult to measure and simple relationships, such as tortuosity = 1/porosity, are often used. The tortuosity can be directly calculated from PFG NMR diffusion measurements and this work investigates the relationship between tortuosity and porosity for a range of alumina and titania pellets and shows that whilst a simple reciprocal relationship between tortuosity and porosity provides a reasonable estimate for the titania pellets, this relationship does not hold for the alumina pellets. This highlights the need of experimental techniques to measure the tortuosity. PFG NMR tortuosity measurements have been extended to catalyst materials containing high concentrations of paramagnetic species. PFG NMR measurements are usually carried out at high magnetic field strengths (>1 T) but this is not possible when samples contain high concentrations of paramagnetic species as the NMR signal lifetime becomes very short. Short signal lifetimes are due to paramagnetic species causing large internal magnetic field gradients which scale with the external magnetic field strength. In this work it is shown that if PFG NMR is performed at low field (2 MHz), the tortuosity of catalyst pellets containing industrially relevant concentrations of paramagnetic species (20 wt.% Co₃O₄/TiO₂ used for Fischer-Tropsch Synthesis) can be measured successfully. Spatially resolved measurements of the NMR signal intensity have been obtained at high resolution (typically 10μm × 39 μm) of a range of titania pellets, some of which revealed significant spatial heterogeneity. These heterogeneities are attributed to spatial differences in the oxidation state of titanium. Ti⁴⁺ is diamagnetic, whereas Ti³⁺ is paramagnetic. These spatial variations could have been introduced by the catalyst production process and could be important in catalyst performance where the presence of Ti³⁺ can influence metal-support interactions. Spatially resolved measurements of diffusion and NMR relaxation parameters (T₁, T₂) have been obtained at high resolution using MRI pulse sequences specifically designed to investigate spatial heterogeneities in catalyst pellets. The imaging methods have been applied to a range of both titania and alumina catalyst support pellets in the shape of trilobes. NMR relaxation measurements of liquid confined in porous material are sensitive towards the surface to volume ratio of the pores, the liquid-solid adsorption strength and the presence of relaxation sinks (including paramagnetic species) at the pore surface. T₂ relaxation maps revealed significant heterogeneities. An increase in T₂ values was for example observed at the pinch points of the trilobes, which can be attributed to a difference in solid-liquid interaction at the pore surface. These heterogeneities could be a result of imperfections in the extrusion or drying process. This thesis demonstrates how NMR methods can be used to gain a more complete and realistic understanding of catalyst pellets and to optimize the manufacturing processes of catalyst pellets.

Ultra-high field (UHF, B0 ≥ 7 T) magnetic resonance imaging (MRI) scanners allow more sensitive, faster and higher resolution imaging than clinical MRI systems at 1.5 T and 3 T. However, radiofrequency field (B+ 1) inhomogeneities can cause signal dropouts in UHF MRI. The most flexible solution to improve image homogeneity is parallel transmission (pTx). A pTx-enabled scanner has multiple independently-driven transmit channels, providing additional degrees of freedom for improved spatial and temporal control of the B+1 fields. This thesis aims to extend the methodology and applications of pTx for neuroimaging at 7 T, specifically on the first clinically-approved model of UHF whole-body 7 T MRI scanner (MAGNETOM Terra). First, this thesis sets out to understand the pTx framework on the new scanner (Chapter 4) and compares the performance of the vendor-provided single transmission head coil against its pTx counterpart (Chapter 5). Image quality is comparable for both coils, with the pTx coil showing significant increases in B+1 values (between 4.9 - 18.9%) in the caudate, insula, putamen, and temporal lobes. This translates into better spinal cord visualisation in MP2RAGE images and reductions in signal dropouts in the temporal lobes for diffusion tensor imaging. Overall, running standard neuroimaging protocols on the pTx coil is reasonable, which then gives opportunity to add dynamic pTx sequences with minimal disruption. Additionally, pTx spokes pulses are incorporated into a gradient-echo echo-planar imaging (GRE-EPI) sequence (Chapter 6). A task-based functional-MRI (fMRI) experiment shows that data quality (expressed by temporal SNR) improves by up to 11% in regions across the whole brain for pTx-EPI in healthy volunteers. The data quality improvement leads to higher median z-scores in two functional paradigms. This sequence has been used for applied fMRI studies at the Wolfson Brain Imaging Centre. Finally, pTx is used to improve 7 T diffusion imaging (Chapter 7). Several aspects are explored: pTx pulses to improve image homogeneity; readout-segmented EPI to reduce T*2 decay and blurring; and various lipid suppression modules are tested to minimise artifacts. Diffusion images with spokes pulses are acquired for single-shot and readout-segmented sequences at 1.5 mm3 isotropic and 0.8 × 0.8 × 3 mm3 resolution, respectively. Furthermore, multi-band 2 readout-segment diffusion images are acquired with B1-shimmed pulses. Overall, this work shows the improvements from pTx in two crucial neuroimaging sequences. There is growing consensus that pTx will become a core technology in UHF imaging, and hopefully, the methods proposed here will help pTx to transit into mainstream neuroimaging.

Traumatic brain injury (TBI) constitutes a major public health problem. The advent of collaborative neuro-imaging studies, advances in magnetic resonance imaging (MRI) and development of serum protein biomarkers have created an opportunity to gain new insights into the recovery after TBI. This thesis aimed to understand the role of MRI and serum biomarkers for outcome prediction after TBI. Before this clinical question could be answered, two methodological challenges had to be resolved. First, outcome data is often missing when patients are lost to follow up. We conducted a systematic review and created a novel framework for the statistical handling of missing outcome data in TBI. Second, MRI data collected at different sites may require harmonization. We examined the effect of different harmonization methods on a variety of imaging parameters, to inform our choice of method for subsequent chapters. The next two chapters concentrated on patients with mild TBI, as they represent the largest proportion of TBI patients and are in great need for better prognostic models. We demonstrated that structural white matter changes are detectable with advanced MRI, including diffusion tensor imaging (DTI), acutely after mild TBI even when conventional MRI is normal, and that these changes correlate with symptoms and recovery. We therefore assessed whether the association between DTI and outcome withstands the scrutiny of a formal prognostic study design and found that DTI substantially improves exiting prognostic models. We next focused on patients with moderate-severe TBI in whom the low level of consciousness is incompletely explained by CT appearances but might be explained by injury detectable with MRI or serum biomarkers. Unlike in mild TBI patients an MRI transfer in such critically ill patients constitutes a clinical risk. We show that protein biomarkers can help triage patients with moderate-severe TBI for MRI. We proceeded to investigate whether serum biomarkers could be used for prognosis directly, without the need for MRI. We found that serum biomarkers did significantly improve established prognostic models in this patient group with discordant CT appearances and conscious level. To conclude, this thesis presents methodological advances in the study of TBI, shows how DTI can transform outcome prediction after mild TBI and how serum biomarkers can improve MRI triage and prognosis in certain patients with moderate-severe TBI.

Introduction: Cerebral small vessel disease (SVD) is increasingly recognised as a key factor in dementia and cognitive impairment, although its role in the pathogenesis of Alzheimer's disease (AD) remains poorly understood. Through the investigation of how SVD relates and interacts with (1) molecular AD pathologies and (2) early midlife risk factors of dementia, this body of work aims to advance our understanding of SVD in relation to AD. Methods: In two separate cohorts, SVD was quantified on 3T MRI through the assessment of white matter hyperintensities (WMH), enlarged perivascular spaces (EPVS), cerebral microbleeds (CMB), and lacunes. Regional data on SVD location was used to derive composite scores for global SVD burden, and the SVD subtypes of hypertensive arteriopathy and cerebral amyloid angiopathy. The first objective was investigated in the Neuroimaging of Inflammation in Memory and Other Disorders (NIMROD) study, which consisted of multimodal PET imaging of older adults with varying degrees of clinical severity. Here, we investigated whether SVD was associated and interacted with the hallmark AD pathologies of amyloid, tau, and neuroinflammation (microglial activation) - these were quantified with PET imaging of [11C]PiB, [18F]AV-1451, and [11C]PK11195, respectively. For insights into the pre-clinical stages of AD pathogenesis, we examined cognitively healthy midlife adults from the PREVENT-Dementia study. We investigated whether well-established risk factors of late-life dementia were related to SVD burden much earlier on, at midlife. Results: SVD was associated with PET markers of neuroinflammation and tau, and plasma Aβ. Notably, associations with amyloid and tau markers were largely limited to WMH. In healthy middle-aged adults, inherited predisposition to dementia (APOE4, parental family history of dementia) was related to longitudinal SVD progression, but not baseline SVD severity. On the other hand, modifiable risk factors of late-life dementia were associated with midlife SVD burden. The two SVD subtypes were differentially related to AD pathologies and risk factors - neuroinflammation and modifiable risk factors were more closely associated with hypertensive arteriopathy, while amyloid and tau related more closely to cerebral amyloid angiopathy. Conclusions: Together, our findings implicate SVD in the early pathogenesis of AD and suggest that lifestyle modifications as early as midlife may reduce the vascular-driven component of AD. Findings highlight the need to further inspect the biological underpinnings of WMH and fluid markers of AD pathology.

Development of efficient pulse sequences for hyperpolarised 13C imaging: Dynamic nuclear spin polarization of 13C-labeled cell substrates has enabled dynamic measurements of tissue metabolism in vivo using 13C MRSI. The principal limitation of the technique is the short lifetime of the hyperpolarisation, which requires the use of very fast imaging methods. As well as being fast the imaging pulse sequence must make economical use of the polarisation since each excitation pulse results in depletion of the polarisation, in addition to that due to T1- dependent decay, which degrades the SNR and decreases the time window over which metabolism of the labelled substrate can be monitored. A single shot 3D multi spin echo pulse sequence with optimised RF pulses was designed in response to these requirements with the aim of clinical translatability. Monitoring tumour cell death using 2H magnetic resonance spectroscopy and spectroscopic imaging: 2H magnetic resonance spectroscopic imaging (MRSI) has been shown recently to be a viable technique for metabolic imaging in the clinic. The relatively low sensitivity of 2H detection is compensated by its very short T1, which means that signal can be acquired rapidly without saturation. Production of [2,3-2H2]malate, following injection of [2,3- 2H2]fumarate (1g/kg) into tumour-bearing mice was measured in various tumour models using surface-coil localized 2H MR spectroscopy at 7T. Malate production was also imaged in EL4 tumours using a fast 2H chemical shift imaging sequence. The possibility of detecting tumour cell death in vivo by means of monitoring [2,3-2H2]malate production from [2,3-2H2]fumarate was demonstrated. As a related pulse design problem, the possibility of reducing the power requirement of the RF pulses used in such experiments by means of numerical optimal control was investigated. Genetic algorithm-based pulse sequence optimisation: The performance of pulse sequences in vivo can be limited by fast relaxation rates, magnetic field inhomogeneity and non-uniform spin excitation. A simple method for pulse sequence optimisation is presented that uses a stochastic numerical solver which, in principle, is capable of finding a better local optimum. The method provides a simple framework for incorporating any constraint and implementing arbitrarily complex cost functions. Efficient methods for simulating spin dynamics and incorporating frequency selectivity are also described. The proposed technique is evaluated with common pulse design problems where a high-quality solution is not yet readily available, such as robust heteronuclear polarisation transfer experiments, excitation pulse design with special requirements e.g. insensitivity to B1-inhomogenity while maintaining a low B1- amplitude and rapid water suppression. The aim was to minimize pulse amplitudes so as to keep the deposited energy as low as possible, which would then allow clinical translation.

Aphasia is a prevalent and debilitating consequence of stroke that often persists chronically. The overarching purpose of this thesis was to better understand the neural compensatory mechanisms that take place to minimise language dysfunction, and promote language recovery, in post-stroke aphasia. This thesis proposes that existing theories of aphasia recovery can be conceptualised as specific examples of two more fundamental principles, degeneracy and variable neurodisplacement (Chapter 2). 'Degeneracy' predicts that upregulated compensatory regions should not be engaged during language in health, while 'variable neurodisplacement' predicts that such regions might be downregulated during health to save resources but upregulated when task difficulty is increased in controls. The thesis then investigates the neural regions that are functionally involved in language recovery post-stroke. Chapter 3 reports an Activation Likelihood Estimation meta-analysis of coordinate-based language functional neuroimaging studies. The language network is bilateral in aphasia and controls. Regions of the right anterior insula and inferior frontal gyrus were more likely to be activated during language in aphasia than controls, overlap with the Multiple Demand network and were more likely to be activated during higher than lower demand language tasks, consistent with enhanced utilisation of spare capacity within right hemisphere executive regions via variable neurodisplacement. Unexpectedly, Chapter 3 found that multiple undamaged midline and right hemisphere regions were less likely to be activated during language in aphasia than controls, consistent with functional diaschisis. Chapter 4 reports one of the first language multivariate pattern analysis functional imaging experiments in post-stroke aphasia and controls. Chapter 4 suggests the existence of a novel form of 'information diaschisis' in which having a stroke is associated with lower language information processing in a bilateral set of undamaged, predominantly domain-general regions which in turn is associated with, and might contribute to, language impairment post-stroke. Chapter 5 analysed longitudinal functional imaging and neuropsychological data from participants with aphasia poststroke. Language profiles were multidimensional at 2 weeks post-stroke and could be represented by three orthogonal components representing fluency, semantic/executive function and phonology. Different language components had uncorrelated recovery trajectories that were associated with changing activation in different neural regions during aphasia recovery. This provides insights into the multidimensional nature of aphasia recovery and suggests that future clinical rehabilitation trials should take a 'personalised medicine' approach that accounts for each patient's specific language profile when deciding targets for non-invasive brain stimulation.

Hypertension is a highly prevalent condition and, as a risk factor for vascular disease in particular, a leading contributory cause of death worldwide. Recent consensus guidelines suggest that moderate and severe (grade II and III) hypertension should be treated rapidly to achieve targets though, prior to the inception of this thesis, the evidence for the safety and efficacy of this approach, together with the physiological consequences of rapid hypertension treatment in moderate and severe disease, was limited. This thesis explores the clinical consequences of an 18-week treatment programme for individuals with grade II and III hypertension, using guideline- recommended pharmacological treatment, delivered over an accelerated timeframe. The blood pressure response to treatment is reported, together with the tolerance and safety of the protocol, as defined by the protocol completion rate, frequency of medication side effects and clinically significant adverse events. The programme also provided an opportunity to study health-rated quality of life in patients with moderate and severe hypertension and the effect of rapid treatment on health-related quality of life. This allowed for the first validation (according to modern standards) of an English language disease-specific instrument for measuring health-related quality of life in hypertension, following translation of the original MINICHAL disease-specific instrument from the original Spanish. In addition, the clinical treatment programme provided an opportunity to study the microvascular response to rapid treatment of moderate and severe hypertension, particularly with relevance to the rarefaction of hypertension and its reversal with treatment. Moreover, the morphological and functional myocardial consequences of treatment were determined, using cardiac MR imaging. Accordingly, this thesis presents evidence supporting the rapid treatment of moderate and severe hypertension, providing an opportunity for this to be studied in future investigations, with the aim of exploring whether this approach is prognostically advantageous for patients.

Multiple sclerosis (MS) is an immune-mediated, neurodegenerative disease of the central nervous system (CNS) characterised by focal demyelination (lesions) and axonal loss. Clinical presentation is heterogeneous; yet the relapsing-remitting MS (RRMS) sub-type is defined by acute clinical episodes, which distinguish it from continuously worsening disability in progressive sub-types. Early intervention with disease modifying therapies (DMTs) may prevent neurodegeneration. Current DMTs, however, generally target inflammation, are expensive and are limited by accompanying adverse effects. In vivo biomarkers are thus warranted for early stratification of patients, longitudinal observation of therapeutic response and evaluation of novel treatments. Magnetic resonance imaging (MRI) plays a key role in MS diagnosis; yet the predictive power of conventional MRI biomarkers, such as T2-w lesion load and brain atrophy, to determine prognosis and disease severity in early disease is weak. Advanced MRI techniques such as magnetisation transfer imaging (MTI) and diffusion-weighted imaging (dMRI), which probe tissue microstructure non-invasively, may be more sensitive to MS neuropathology, and show promise as putative biomarkers of underlying MS pathology. Quantitative MTI has been widely applied in MS studies. Poor intra- and inter-site reproducibility of the derived magnetisation transfer ratio (MTR) metric and the time-consuming protocols required for fully quantitative MTI have restricted clinical usage of MTI. MTsat, a semi-quantitative MTI measure, is less influenced by variance in the MTR signal caused by T1 relaxation and B1 inhomogeneities, and thus may be a better prognostic biomarker than MTR. Combining quantitative MTI with dMRI measures may also improve specificity to the heterogeneous biological processes which occur in MS at diagnosis. In this work, I assessed MTI literature in RRMS through systematic review and meta-analyses. I compared the reproducibility of MTsat with MTR in healthy volunteers. I also assessed the reproducibility of the MRI aggregate g-ratio, derived from combined MTsat and dMRI-derived neurite orientation dispersion and density imaging (NODDI) measures. I validated the use of MTsat in a clinical cohort of people recently diagnosed with RRMS, before therapeutic intervention, and evaluated the relationship between MTI biomarkers and clinical disability. We substantiate the MRI g-ratio as a biomarker in RRMS by comparing it with an established blood biomarker of axonal damage, neurofilament. Finally, I determine the longitudinal evolution of MTsat, compared with MTR, and g-ratio in early RRMS.

The diagnosis of Creutzfeldt-Jakob disease (CJD) is dependent on an assessment of clinical features together with specialist investigations including electroencephalogram (EEG), magnetic resonance imaging (MRI) of the brain and the cerebrospinal fluid (CSF) tests 14-3-3 immunoassay and real-time quaking induced conversion (RT-QuIC). There has been an evolution in the diagnosis of CJD with time as new tests have been developed and this includes MRI, which has proved very useful in the diagnosis of all forms of CJD, including sporadic CJD (sCJD) and variant CJD (vCJD). MRI signal abnormalities in sCJD have a relatively high diagnostic sensitivity and specificity, although this varies in different studies in relation to the precise details of the study population. The primary aim of this thesis: To examine the overall role of MRI brain scan in the diagnosis of sCJD. Other, specific, aims: -How accurate is local reporting of the MRI findings in CJD and has this changed with time? -Is there variation in the utility of MRI in the diagnosis of CJD according to codon 129 genotype and prion protein isotype -Is there evidence that MRI in sCJD and vCJD is more likely to show characteristic features at different stages of the clinical illness and what is the role of serial scans? -Is there a relationship between specific MRI appearances and the clinical presentation in sCJD? -Is the MRI of use in identifying cases of methionine-valine (MV) vCJD? -Are there scans in non-CJD cases that show features suggestive or characteristic of CJD, and are these cases distinguishable from true cases on the basis of clinical features or other investigations? -Are there cases of CJD in which the scans show no characteristic features, and are there any parameters that distinguish these cases, for example scan quality or timing? In order to assess the study aims, 462 cases of sCJD referred the National CJD Research and Surveillance Unit (NCJDRSU), identified as having an MRI brains scan performed during their illness, were analysed retrospectively (2010-2015). 54 non-CJD cases (referred to NCJDRSU, with an initial suspicion of CJD, but who turned out to have other illnesses) were included for comparative analyses (2010-2015). In addition, 25 cases of vCJD with brain MRI DWI sequences (1999-2015) were identified and these scans were analysed. The scans were reviewed by a single Neuroradiologist, experienced in the assessment of MRI, for the presence or absence of features associated with these diseases. The Neuroradiologist was aware of the suspicion of CJD but was blinded to the specific diagnosis in each case. This thesis determines the role of MRI brain scan in the diagnosis of sCJD and provides further validation of the current European diagnostic criteria (2017) for the condition. The results show that brain MRI, with the inclusion of DWI sequences, is a highly sensitive, specific and reliable test for the diagnosis of sCJD if applied in the appropriate clinical circumstances and can improve the diagnostic classification of sCJD in life.

Image processing and analysis techniques have become more prevalent within the medical field as they aid in the detection, diagnosis and prognosis of conditions that threaten people's lives. Some conditions, however, present additional challenges to medical experts, requiring a further and more advanced examination of the patient's data before a decision can be made. The use of automated systems can aid in the task as they provide objective tools to the experts to support their analysis. The work presented in this doctoral thesis contributes to this field by exploring a novel approach within the field of representation learning for the processing and analysis of images. For this, we focus on the diagnosis of Alzheimer's Disease (AD) from Magnetic Resonance Imaging (MRI) scans. Our approach follows that of the conventional machine learning techniques, in which the task is split into two stages, one devoted to extract the features from the data whilst the second one fits a classification model. The main contribution of our work lies in the first stage. For this, we exploit a generative model describing images by making a series of assumptions that we hypothesised should improve the task of interest. The generative models that we propose fall within the sparsity-based category and we consider in particular Dictionary Learning and Sparse Coding. We further expand our approach by carrying the initial assumptions to the Convolutional Sparse Coding realm. The standard approaches that exploit sparsity rely on the optimisation of a cost function with a regulariser based on the L1- or the L0-norms. In our work, however, we assume the sparse coefficients are modelled by the Cauchy distribution, which results in a new penalty term to be optimised during training. This new regularisation term is exploited as the core of our encoding algorithm and we initially demonstrate its power for the tasks of reconstructing 2D images. We ultimately make use of the codes learnt with this approach for the task of interest: the diagnosis of Alzheimer's Disease (AD). For this we learn and extract features from the MRI scans that serve as input for a classifier, a Fully Connected Neural Network (FCNN). The results obtained are competitive with the state-of-the-art and encourage further exploration and expansion of the proposed approach. Our work is benchmarked against a deep Convolutional Neural Network (CNN), one of the state-of-the-art techniques in computer vision for biomedical applications.

Introduction: Heavy menstrual bleeding (HMB) is a common and debilitating condition. Current treatment options are limited by lack of effectiveness, side effects and surgical risks. Better understanding of the physiology of menstruation may help identify new therapeutic targets. Transient vasoconstriction of the endometrial spiral arterioles occurs prior to menstruation. There is in vitro evidence that endometrial hypoxia is present at menstruation and activates endometrial repair and that a decreased hypoxic response may result in HMB. Determination of hypoxia in vivo has not yet been possible in the human endometrium. Magnetic resonance imaging (MRI) techniques may be utilised to non-invasively detect markers of endometrial hypoxia in women. Obesity, defined as a body mass index (BMI) ≥30 kg/m2, has a profound impact on health and its prevalence is rising worldwide. However, there are limited data on the influence of obesity on menstrual blood loss (MBL). Therefore, it was hypothesised that: 1.Hypoxia is present in the human endometrium at menstruation and can be detected in vivo and ex vivo.2.Women with objectively measured HMB have a defective endometrial hypoxic response during menstruation.3.MRI assessments of endometrial hypoxia correlate with laboratory-based techniques in the detection of hypoxia markers. 4.Obesity in women is associated with increased MBL due to an impaired endometrial hypoxic response at menstruation. Methods: 24 participants were recruited and underwent measurement of MBL using the modified alkaline haematin method (HMB: ≥80 ml) and attended for MRI scans timed for the menstrual and early/mid-secretory phase of the menstrual cycle. To address hypotheses 1-3, the protocol included T2* quantification and dynamic contrast-enhanced MRI (DCE-MRI). An endometrial sample was also collected in the menstrual and early/mid-secretory cycle phase for laboratory analysis of hypoxia markers: CAIX (a hypoxia marker) analysed by immunohistochemistry (IHC) and hypoxia-regulated repair factors (ADM, VEGFA, CXCR4) and lactate dehydrogenase (LDHA) measured by quantitative reverse transcription polymerase chain reaction (qRT-PCR). Endometrial vasculature was also analysed by immunofluorescence for CD-31 (endothelial cell marker) and α-SMA (marker of vascular smooth muscle cells).To assess hypothesis 4, women (n=122) completed a pictorial bleeding assessment chart (PBAC) and had their BMI measured. Endometrial samples were also collected from 28 women in the menstrual/late-secretory cycle phase to assess for hypoxic markers as above. Results: In vivo imaging revealed women with normal menstrual bleeding (NMB; < 80 ml) had reduced endometrial T2* at menstruation compared to the non-menstrual phase. Deoxyhaemoglobin increases local magnetic field inhomogeneity which decreases T2*, hence these data are consistent with hypoxia at menses. Endometrial plasma flow (Fp) was increased at menstruation compared to the non-menstrual phase which may reflect post hypoxic vasodilation of the spiral arterioles within the endometrium. Ex vivo findings in women with NMB showed increased levels of ADM, VEGF-A, CXCR4 and LDHA mRNA in paired menstrual compared to non-menstrual endometrium, supporting the presence of endometrial hypoxia at menstruation. Comparing women with NMB and HMB, in vivo imaging showed no difference in T2* at menstruation. However, endometrial Fp and plasma volume (vp) were reduced at menstruation in women with HMB compared to NMB. This may reflect delayed initiation of endometrial repair or a lack of post hypoxic vasodilation within the endometrium of women with HMB. There were no differences in laboratory analyses of endometrial hypoxic markers at menstruation between these two groups of women. A significant negative correlation between T2* and endometrial concentrations of ADM, VEGF-A, CXCR4 was observed providing validation of this non-invasive technique for the detection of endometrial hypoxia in vivo. There was a positive correlation between BMI and MBL but no evidence that obesity impairs endometrial hypoxia at menstruation. In summary, the novel in vivo and ex vivo data presented in this thesis support the presence of hypoxia in the endometrium of women with NMB. Women with HMB displayed altered endometrial perfusion at menstruation. Non-invasive imaging may improve diagnosis and provide opportunity for identification of personalised targets for the management of HMB.

There has been a global rise in migration due to factors including conflict, famine, economic deprivation, health inequalities and more. This has resulted in an increase in the necessity for age assessments to be conducted. This requires methodologies that are both scientifically robust and ethical. Clinical imaging is generally the approach of choice and Magnetic Resonance Imaging (MRI) is of increasing value as it permits visualisation of osteological maturation which is strongly correlated with chronological age and it meets necessary health and safety requirements. However, limitations exist in the reproducibility and comparison of studies based on MRIs. To identify accurate and appropriate age estimation methodologies this research project aimed to chart maturation of the distal end of the femur using T1 weighted MRIs, relate the changes to chronological age and develop an age estimation method that could be used to predict the age with accuracy and repeatability. A data set of 321 (164 males,157 females) T1 weighted MRIs were collected from a Scottish medical sample aged between 10 and 20 years at the time of image acquisition. An eight-stage classification system was developed and maturational maps of the femoral growth plate were created for each MRI and for each 1-year age cohort. To facilitate comparison of the maturational maps, 8 areas within the created maps were identified and tested to establish whether significant differences existed between them. To relate maturational stage to chronological age, age ranges for each of the 8 stages within the classification system were obtained. Twelve machine learning random forest and random forest with hyperparameter grid regression models (6 for male, 6 for females) for age prediction using 6 different combinations of feature factors were developed. The models were then tested on 40 MRIs as proof of concept. From the results it was observed that the growth plate starts to fuse in the central area and fusion progresses centrifugally. The results showed that when utilising MRIs as the imaging modality the first bone bridges were present in females by 11 years 1 month (133 months) and in males by 12 years 4 months (147 months). The growth plate was mature by 15 years 4 months (184 months) in females, and 16 years 9 months (201 months) in males. Complete fusion of the growth plate at the distal end of the femur can be of assistance in determining the minimum age of 16 years in males, and 14 years in female individuals. It was not possible to assess the minimum age of 18 years in neither sex. A series of stages along with their location within the growth plate that can be useful for age prediction were also identified, specifically for the age thresholds of 12, 14, 16 and 18 years. From the analysis of the age prediction models, the random forest with hyperparameter grid gave the best results. For females the best model results gave a mean absolute error (MAE) of 12.73 months, R2 0.79, while for males a MAE of 9.20 months, R2 0.88. When considering location of the stages within the growth plate when predicting age, the age assessment did not improve, and when using only selected areas of the growth plate the MAE in months increased, suggesting that the whole growth plate should be used when predicting age, instead of selected images or areas. In conclusion, maturation of the distal end of the femur was described for an age cohort of 10-20 years of age, providing final recommendations to assess age from MRIs of the distal end of the femur for relevant age thresholds.

Introduction: Schizophrenia symptomatology exists on a continuum ranging from subclinical psychotic-like experiences in the general population (schizotypy) to full-blown clinical symptoms. High schizotypy individuals are at an increased risk for developing clinical diagnoses, yet previous work has not investigated key neural abnormalities of schizophrenia in these samples. Methods: To ensure findings are informative for clinical risk, we only recruited individuals scoring at the extreme low/high end of the Schizotypy Personality Questionnaire. 27 high schizotypy (HS) and 26 low schizotypy (LS) individuals to take part in two functional magnetic resonance imaging (fMRI) tasks assessing social learning. Participants also underwent a resting state fMRI and a magnetic resonance spectroscopy scan. Results: HS subjects, compared to LS, present with abnormal learning of social information. HS overestimate the volatility of social cues and are slower to learn about global changes in social context. Furthermore, HS subjects show reduced neural activity in the dopaminergic midbrain and increased frontal cortex activity in response to prediction errors during social learning. HS subjects also present with a reduced resting-state functional connectivity between hippocampus and striatum/thalamus and with reduced GABA and Glu metabolite levels in the prefrontal cortex. Discussion: HS subjects, representing the earliest risk for clinical transition to schizophrenia, already present with key neural abnormalities implicated in progression, mainly abnormal hippocampal functioning and abnormal GABA/Glu levels. These results encourage investigations of HS to facilitate a comprehensive view of risk/protective factors for clinical transition. The results also show that HS subjects already present with abnormal hierarchical learning as seen in clinical samples. HS subjects neutrally underweight prediction errors indicating an improper processing of these learning cues. They also present with compensatory activity in frontal cortex enabling behavioural performance similar to LS. The abnormal learning from social cues could explain not only the social functioning deficits key to schizophrenia, but also other cognitive biases observed in these populations.

The Fischer-Tropsch (FT) reaction is a heterogeneous catalytic reaction that produces synthetic linear hydrocarbon fuels from syngas. Recently there has been renewed academic and industrial interest in the reaction due to its potential to be used in the production of carbon-neutral fuels. The reaction typically occurs inside mesoporous catalyst particles. It is well known that at steady state, the catalyst mesopores are filled with heavy liquid hydrocarbon product. The diffusion of reactants and products through this liquid has a significant influence on reaction selectivity and activity, however this is currently poorly understood in the literature. This thesis presents the use of magnetic resonance techniques to understand this diffusion, and in particular how it depends on the composition of the heavy liquid product. The Cohen-Turnbull-Bueche (CTB) free volume model of diffusion is shown to be an accurate description of the diffusion of liquid hydrocarbon products. As such, a novel optimisation method is developed that is able to calibrate the model when applied to hydrocarbon mixtures confined in mesopores. This enables the reconstruction of carbon number distributions of confined mixtures using Pulsed-Field Gradient (PFG) Nuclear Magnetic Resonance (NMR) techniques. An advancement on this calibration technique is used to show that the diffusion mechanism of these liquids, in addition to the activation energy and molecular free volume, changes significantly when they are confined in mesopores compared to their bulk liquid counterpart. Further, temperature variations at FT conditions are shown to influence the diffusion mechanism. It is also found that a simple Arrhenius activation energy description of mixture diffusion fails to yield the correct enthalpic activation energy; a more careful treatment based on the CTB model yields the correct result. The CTB model is parametrised for FT reaction conditions at 220 ºC. Using spatially resolved PFG NMR, the carbon number distributions of liquid products as they form in situ inside catalyst mesopores during an operando FT reaction have been computed. Resolution of pores within a single catalyst pellet was achieved. It is shown that the average carbon number of liquid products inside pores is smaller than liquid products accumulating between pellets. Finally, an analysis of the diffusion of CO and H2 (syngas) dissolved in model FT hydrocarbon liquid confined in mesopores is presented. Using the results, a simplified model based on existing literature that depends only on the solute and the temperature is shown to describe well the temperature dependence of diffusion.

Adolescent major depressive disorder (MDD) can have devastating consequences that remain into adulthood. Although the severe effects on those that suffer from MDD are well understood, the neural mechanisms underpinning the illness are still unclear, as the past literature exploring adolescent MDD has suffered greatly from a lack of consistency. In particular, it is still unclear as to what deviations of brain structure and function actually occur in adolescent MDD. Furthermore, though much of the past literature has claimed that cognitive behavioural therapy (CBT) has a normalising effect on disrupted brain function in adolescent MDD, this has not been directly tested. Additionally, past literature has found that adult MDD patients with a high familial loading for MDD, defined as having at least one first degree relative with MDD, show different symptom profiles to MDD patients with a low familial loading for the illness. However, it is not yet clear whether these symptom differences also occur in adolescent MDD patients with differing familial loadings, nor is it clear whether these two patient groups also differ with respect to brain structure and function. If so, familial loading for MDD could potentially be used to subtype adolescent MDD patients. This thesis aimed to further elucidate the neural processes that may be unfolding within the depressed adolescent brain by investigating pre-treatment differences in cortical thickness, white matter volume, and resting-state functional connectivity, using the largest studied sample of adolescent MDD patients under the age of 18 years. This thesis further aimed to investigate CBT's effects on brain function, and whether adolescent MDD patients could be subtyped based on familial loading for MDD. Before receiving treatment, adolescent MDD patients showed structural deviations in the form of greater cortical thickness and white matter volume within frontal and limbic regions of the brain. Furthermore, adolescent MDD patients also showed extensive pre-treatment overconnectivity within multiple functional brain networks, being the fronto-limbic, default mode, central executive, and salience networks. Moreover, when investigating the CBT-related changes to resting-state functional connectivity, in adolescent MDD patients, and how they relate to regions showing pre-treatment functional disruption, it was found that regions showing the greatest pre-treatment functional disruption actually showed the weakest CBT-related changes in resting-state functional connectivity. This suggests that CBT does not have a normalising effect on brain function in adolescent MDD but instead may have a rehabilitative effect on resting-state functional connectivity. Finally, when separating adolescent MDD patients into those with a high or low familial loading for MDD, the two patient groups differed with respect to the structure and function of the default mode network, as well as differing in rumination symptoms. Together, this work demonstrates the need for larger sample sizes to be used when investigating an illness as heterogenous as adolescent MDD, as well as direct investigations into the potential mechanisms of CBT. Moreover, it appears that some of this heterogeneity found in adolescent MDD may be partly addressed by subtyping patients, which may have implications for how the illness is viewed and treated in the future.

Immune checkpoint inhibition using monoclonal antibodies targeting cytotoxic T-lymphocyte antigen-4 (CTLA-4) and programmed cell death receptor-1 (PD-1) to enhance cytotoxic T-lymphocyte response against tumours has been used in the management of several malignancies including metastatic melanoma. Although preliminary studies in patients with untreated metastatic melanoma demonstrated significantly increased progression-free survival, there is an unmet need for clinical decision-making tools that can be used for monitoring the efficacy of these costly agents, their potential side effects and the identification of patients with the most suitable immunological profile for this treatment approach. The development of non-invasive imaging methods and imaging biomarkers of response to immune checkpoint inhibitors could therefore provide important diagnostic and prognostic information. This project builds on existing technologies in Cambridge to develop structural and functional imaging approaches for monitoring tumour response to immune checkpoint inhibitors. Magnetic resonance imaging (MRI), radiolabelling of neutrophils for single-photon emission computed tomography (SPECT) imaging and intravital imaging are explored as part of a co-clinical study on metastatic melanoma patients and syngeneic mouse models of cancer immunotherapy.

Neuroimaging studies are becoming increasingly bigger, and multi-centre collaborations to collect data under similar protocols, but different scanning sites, are now commonplace. However, with increasing sample size the complexity of databases and the entailed data management as well as computational burden are growing. This thesis aims to highlight and address challenges faced by large multi-centre magnetic resonance imaging(MRI) studies. The methods implemented are then applied to traumatic brain injury (TBI) data. Firstly, a pre-processing pipeline for both anatomical and diffusion MRI was proposed, that allows for a high throughput of MRI scans. After describing the choices for processing tools,the performance of the integrated quality assurance was assessed based on the results from a large multi-centre dataset for TBI. Secondly, the applicability of the pipelines for processing mild TBI (mTBI) data from three sites was shown in a case study. For this, volumetric and diffusion metrics in the acute phase are analysed for their prognostic potential. Further-more, the cohort was examined for longitudinal changes. Thirdly, independent scan-rescan datasets are examined to gain a better understanding of the degree of reproducibility which can be achieved in imaging studies. This involves analysing the robustness of brain parcellations based on structural or diffusion imaging. The effect of using different MRI scanners or imaging protocols was also assessed and discussed. Fourthly, sources of diffusion MRI variability and different approaches to cope with these are reviewed. Using this foundation,state-of-the art methods for diffusion MRI harmonisation were compared against each other using both a benchmark dataset and mTBI cohort. Lastly, a solution to localise brain lesions was proposed. Its implications for lesion analysis, are assessed in the light of an application to a more severe TBI patient cohort, imaged on two different scanners. Furthermore, a lesion matching algorithm was introduced to automatically examine lesion evolution with time post-injury. In summary, this thesis explored different options for MRI data analysis in the context of large multi-centre studies. Different approaches are studied and compared using a number of different MRI datasets, including scan-rescan data across different MRI scanners and imaging protocols. The potential of the optimised solutions was illustrated through applications to TBI data.

This thesis presents the development and implementation of magnetic resonance techniques to study biopharmaceutical samples, specifically from upstream bioreactors. The motivation for this research is to improve our ability to produce therapeutic proteins of appropriate quality and quantity. The aim of this research is therefore to explore the extent to which NMR techniques can be used to quantify the ratio of intra and extracellular proteins present in a biopharmaceutical production line; with the purpose of building up a quantitative method which can be used to non-invasively monitor quality. Although NMR techniques exist for the study of macro-proteins and in-cell environments, previous studies have mostly failed to design NMR-tools capable of evaluating these samples in their native form, or transferrable to bench-scale spectrometers for at-line industrial application. Bovine serum albumin and an IgG1 antibody are used as model extracellular proteins. Due to the difficulties imposed by the large size of the proteins and the intense resonances produced by their media solutions, the experimental techniques require optimization of the NMR pulse parameters and effective solvent suppression. The new pulse sequences T1-WATERGATE, T2-WATERGATE and Diff-WATERGATE are proposed, which allow for the visualization and assessment of macro-protein signals. The effects of protein concentration on diffusion coefficients, relaxation times and spectral peak area are explored. The possibility of developing calibration curves for concentration determination are shown through both the use of 1D 1H WATERGATE spectra and the water-proton transverse relaxation rate (R2(1H2O)), with the latter being transferrable to bench-scale. Chinese hamster ovary cells were used to provide an intracellular environment, which is found to have an associated self-diffusion coefficients 70 times slower than that of extracellular proteins. This work demonstrates how varying the NMR acquisition parameters allows for the self-diffusion coefficients of proteins of different sizes or of different cellular environments to be obtained. Additionally a novel means of monitoring the growth of a mammalian cell line at bench-scale is shown through simple one-shot CPMG experiments. Lastly, this study presents the first use of magnetic resonance imaging (MRI) velocimetry to study the flow field inside an ambr15® microscale bioreactor. MRI provides a non-invasive method for the highly accurate visualisation of these hydrodynamics and a means for validating CFD predictions.

We report a new synthetic strategy for the preparation of high relaxivity macromolecular MRI contrast agents. Five gadolinium(III) complex-es bearing either one or two polymerisable methacrylamide groups were designed and synthesised, serving as monomers, crosslinkers or chain transfer agent respectively, to generate water-soluble copolymers using RAFT polymerisation. Using this approach, polymeric contrast agents (CAs) were synthesised with different architectures, including linear, hyperbranched polymers and gels. The relaxivities of the polymer CAs were determined by NMR relaxometry.

Ischaemic stroke is a major cause of adult death and disability and is one of the most expensive neurological conditions. Currently, thrombolysis by rtPA is the only pharmacotherapy for hyperacute ischaemic stroke, and in routine practice it must be administered within 4.5 hours of symptom onset. Consequently, most patients with unknown onset time are precluded from thrombolytic therapy, despite the possibility of still being within the 4.5 hour treatment window. Unknown onset time, therefore, poses a substantial challenge in the management of hyperacute stroke patients. Multiparametric quantitative magnetic resonance imaging (qMRI) is a potential tool for estimating stroke onset time due to its sensitivity to hydrodynamic and haemodynamic changes in the ischaemic brain. This thesis explores the potential abilities of MRI based stroke timing methods in a preclinical rat model of ischaemia and hyperacute ischaemic stroke patients. Both the preclinical and clinical results suggest that quantifying the change in the T2 relaxation time is a more accurate and reliable method for onset time estimation than measuring the intensities of signals represented in the T2-based weighted images typically acquired in clinics. Results also suggest that T2 distributional information within grey matter lesions may provide a reference-independent stroke timing method. The translation of preclinical methods and MRI findings to stroke patients demonstrated in the thesis highlights the importance of preclinical research in identifying MRI biomarkers that could aid treatment decisions. Clinical MRI scanners can transform conventional weighted magnetic resonance images that are usually acquired and assessed visually (e.g., T2 weighted), into their ‘absolute’ parametric counterparts (e.g., T2 relaxometry maps). Exploiting this ability will enable ischaemia to be identified and onset time estimated more accurately and reliably. Ultimately, the encouraging results from this thesis warrant further investigation into the quantification of T2 for informing treatment decisions regarding hyperacute ischaemic stroke patients, in a more extensive clinical study.

Introduction: Cerebral small vessel disease (SVD) is responsible for a third of strokes and is the most common cause of vascular dementia. It is characterised by white matter lesions and more diffuse damage outside of the lesions. Despite its importance, the mechanisms causing white matter damage are incompletely understood. Both increased blood-brain barrier (BBB) permeability and inflammation have been suggested to play a role. We hypothesised that BBB permeability and microglia activation would be increased in SVD and explored the relationship between these processes and other imaging, cognitive and blood markers. Methods: A PET/MR system was used to acquire simultaneous dynamic contrast enhanced MRI with Patlak modelling to measure BBB leakage, and [C11]PK-11195 PET to assess microglia activation. We recruited 20 sporadic SVD patients, 20 monogenic SVD patients (CADASIL) and 20 healthy controls to this study. Results: BBB permeability in the white matter of the sporadic SVD group was significantly higher than control (p=0.001), with a significantly higher volume of focal hotspots (p=0.010). The focal hotspots of activated microglia also had a significantly higher mean in the white matter (p<0.001). The role of BBB permeability and PK binding in CADASIL was less clear, with no difference in permeability but a significantly higher mean of focal hotspot of activated microglia in the CADASIL white matter, compared to control (p=0.015). In the sporadic SVD and CADASIL groups, the number of voxels containing both significant permeability and microglia activation was significantly lower than expected (p < 0.001). Conclusion: Areas of focal increase in both BBB permeability and microglia activation occur in SVD and therefore may have a role in the pathophysiology of white matter damage. In this cross-sectional dataset, the two processes were spatially independent of one-another. However, we cannot exclude a temporal relationship, where one process may precede another, without follow-up data. Further research is needed to see if these markers predict white matter damage in longitudinal studies, and whether they can be altered therapeutically.

This thesis examines the potential utility of magnetic resonance (MR) quantitative imaging biomarkers (QIBs) of knee osteoarthritis (OA) for rapid assessment of treatment efficacy in experimental medicine studies. The development of treatments able to modify disease in OA is hampered by an inability to evaluate treatment response over a timeframe relevant to clinical trials. There are particular challenges in the experimental medicine setting due to the small numbers of participants and short follow-up duration relative to the expected time course of OA development and progression. Multiple MR QIBs of OA exist which may help address the problem of early evaluation of treatment response. However, their use in early phase studies has remained limited. Possible reasons for this include incomplete characterisation of the performance of QIBs in this setting and lack of head-to-head comparison of candidate QIBs to determine which would be optimal. This thesis aims to address these shortcomings and provide new information on the likely utility of MR QIBs in the setting of experimental medicine studies, as well as their potential for improving our general understanding of OA pathophysiology. I start by examining the reliability and ability to discriminate between OA and healthy knees of cartilage compositional MR imaging in a systematic review and meta-analysis. I then describe the development and validation of a novel semi-automatic surface-based method for analysing articular cartilage composition and morphology at the knee which may offer improved responsiveness and spatial localisation of change. Moving to QIBs of subchondral bone, I evaluate the association between measures of subchondral bone architecture derived from MR texture analysis and OA progression in the Osteoarthritis Initiative. The remainder of the thesis describes a prospective observational study where the utility of MR QIBs of synovium, subchondral bone and cartilage in experimental medicine studies is assessed. In summary, this thesis will inform decisions regarding the use of MR-based QIBs in future longitudinal and interventional studies. Their inclusion in experimental medicine studies may allow early assessment of treatment efficacy at a structural level and improve efficiency of treatment development pipelines.

Non-invasive neuroimaging methods, such as MRI, provide a window into the structure of the mammalian brain. However, despite the ubiquity of these methods, the biological interpretation of the information obtained using these tools remains elusive. In order to accurately link this macroscale data to microscale measurements, it is critical that the construct validity is high. This thesis provides novel analyses, pipelines and methods to: i) generate and validate maps of brain organisation obtained via MRI, and ii) demonstrate the utility of these methods in capturing elements of cognition and psychopathology. First, in Chapter 1, I review some of the neuroscientific context for the new methods presented, from cytoarchitecture to gene expression to connectomes. Chapters 2-4 introduce a new method, “Morphometric Similarity Mapping”, which captures the brain organisation of an individual by mapping the relationships of multiple features of the cerebral cortex. Chapter 2 focuses on the development of the analysis pipeline and the graph theoretical features of the resulting morphometric similarity networks (MSNs), with an emphasis on reproducibility. Chapter 3 highlights the generalisability of MSNs to the macaque monkey, linking MSNs to ex vivo tract tracing experiments and presenting new tools for processing non-human imaging data; as well as evidence that MSN topography is organised by cytoarchitectonic features. Chapter 4 is focused on determining the transcriptomic correlates of MSNs using publicly available gene expression maps, and on applying MSNs to examine the relationship between brain organisation and intelligence. Chapter 5 is dedicated to rigorous evaluation of the applicability of MSNs to measure specific disease-relevant phenotypes in 8 rare genetic disorder cohorts. This includes the validation of novel methods for utilising data from both single-cell sequencing technologies and differential gene expression experiments (in multiple tissue types) in analysing neuroimaging and bulk transcriptomic brain maps. Chapter 6 provides a brief summary and presents some ongoing and future projects expanding on this original work. It also importantly discusses a general framework of comparing brain maps, including MSNs and gene expression, as well as other canonical maps of brain structure and function. Altogether, this thesis presents and evaluates novel methods and applications for integrating multimodal neuroimaging data with genetic data derived from multiple tissue types and through various acquisition strategies. It also includes tools for performing these analyses in non-human primates, and pipelines for statistically comparing brain maps. These results not only provide insight into the manifestation of brain-related changes due to various components of human variation, but also provides a framework for evaluating this variation at multiple biological scales purely from non-invasive neuroimaging data.

In most cases, spontaneous intracerebral haemorrhage (ICH) is thought to result from cerebral small vessel diseases (SVDs). Cerebral amyloid angiopathy (CAA) and arteriolosclerosis (non-CAA SVD) are the two main types of SVD associated with ICH. The risk of recurrent ICH and post-stroke dementia may be higher with CAA-associated ICH compared with non-CAA SVD-associated ICH. It is, therefore, important to identify the type of SVD associated with ICH. Commonly, ICH in the cerebral lobes (lobar ICH) is thought to be associated with CAA, whereas non-lobar ICH is associated with non-CAA SVD. However, this generalisation is not accurate. The aims of my thesis were to (a) histopathologically assess the types and severity of SVDs associated with lobar and non-lobar ICH using brain autopsy tissue, (b) explore whether brain imaging (magnetic resonance imaging (MRI), positron emission tomography (PET) and computed tomography (CT)) biomarkers can be used in living patients to identify the type of SVDs associated with ICH and (c) determine whether CT biomarkers can predict outcome after ICH. Throughout my thesis, I use data from two overlapping studies of ICH. The Lothian Audit of the Treatment of Cerebral Haemorrhage (LATCH) is a prospective community-based audit of all residents in the Lothian Health board region of Scotland who were aged 16 years or above and had an incident ICH between 1st June 2010 and 31st May 2013 inclusive. During LATCH and until 31st May 2016, consecutive adults with SVD-associated ICH were able to consent to have apolipoprotein E genotyping, brain MRI and brain autopsy as part of the prospective Lothian INtraCerebral Haemorrhage Pathology, Imaging and Neurological outcome (LINCHPIN) study. In an autopsy study of 126 LINCHPIN participants with first-ever SVDassociated ICH, I found that 98% of participants with a non-lobar ICH had moderate or severe non-CAA SVD. In contrast, 60% of participants with a lobar ICH had moderate or severe CAA, while many had moderate or severe non-CAA SVD, either with or without CAA. The most common approach used in clinical practice to identify CAAassociated ICH is the MRI-based modified Boston criteria for CAA. However, I found that only about one-third of ICH patients during LATCH were able to undergo MRI. Also, in 16 LINCHPIN participants with research MRI, the modified Boston criteria showed limited accuracy against the histopathological assessment of CAA on subsequent research autopsy. Molecular imaging using tracers designed to detect parenchymal β-amyloid in Alzheimer’s disease, such as flutemetamol, may also identify the perivascular β-amyloid found in CAA. I found that in LINCHPIN brain tissue samples, the amyloid tracer flutemetamol labelled both parenchymal and perivascular β amyloid. Among 20 participants with first-ever ICH, flutemetamol PET scans were 86% sensitive and 77% specific for CAA-associated ICH based on the modified Boston criteria. Brain CT is the test that usually diagnoses ICH. I found in 62 LINCHPIN participants with first-ever lobar ICH that subarachnoid haemorrhage and finger-like projections on CT, and APOE ε4 allele possession were independently associated with CAA-associated lobar ICH defined on subsequent autopsy. I developed diagnostic models and criteria (“Edinburgh criteria”) based on these predictors that could accurately rule in or exclude CAA-associated lobar ICH. I performed a multicentre external validation study of the Edinburgh criteria. In a preliminary analysis, the CT-only diagnostic criteria (subarachnoid haemorrhage and finger-like projections) had high sensitivity (88%) and good specificity (84%) for CAA-associated lobar ICH. The risk of recurrent ICH is a key outcome for ICH survivors. I found that the risk of recurrent ICH was significantly higher in participants with first-ever lobar compared with non-lobar ICH. Among lobar ICH participants, the risk of recurrent ICH was significantly higher in those classified as high risk on the CT-only Edinburgh criteria for CAA-associated lobar ICH compared with those classified as low risk. Predicting death or disability is also important in ICH. I found that the severity of SVD on the diagnostic brain CT was an independent predictor of death or disability at one year after the ICH, after adjusting for known prognostic factors. Brain imaging biomarkers, particularly on brain CT, are potentially useful for identifying the type of SVD associated with ICH. Having developed the CTbased Edinburgh criteria for CAA-associated lobar ICH, I performed preliminary multicentre studies assessing their diagnostic accuracy and prognostic value. In the future, I aim to perform the full analyses of these diagnostic test accuracy and prognostic studies to determine the clinical utility of the Edinburgh criteria. Other research should assess the clinical and economic effect of the Edinburgh criteria on the prognosis of ICH. More work is needed to investigate the prognostic relevance of coexistence of CAA and non-CAA SVDs in lobar ICH, and whether these groups can be differentiated using imaging biomarkers.

White matter hyperintensities (WMH) are neuroradiological features seen in T2 Fluid-Attenuated Inversion Recovery (T2-FLAIR) brain magnetic resonance imaging (MRI) and have been commonly associated with stroke, ageing, dementia, and Alzheimer’s disease (AD) progression. As a marker of neuro-degenerative disease, WMH may change over time and follow the clinical condition of the patient. In contrast to the early longitudinal studies of WMH, recent studies have suggested that the progression of WMH may be a dynamic, non-linear process where different clusters of WMH may shrink, stay unchanged, or grow. In this thesis, these changes are referred to as the “evolution of WMH”. The main objective of this thesis is to develop machine learning methods for prediction of WMH evolution in structural brain MRI from one-time (baseline) assessment. Predicting the evolution of WMH is challenging because the rate and direction of WMH evolution varies greatly across previous studies. Furthermore, the evolution of WMH is a non-deterministic problem because some clinical factors that possibly influence it are still not known. In this thesis, different learning schemes of deep learning algorithm and data modalities are proposed to produce the best estimation of WMH evolution. Furthermore, a scheme to simulate the non-deterministic nature of WMH evolution, named auxiliary input, was also proposed. In addition to the development of prediction model for WMH evolution, machine learning methods for segmentation of early WMH, characterisation of WMH, and simulation of WMH progression and regression are also developed as parts of this thesis.

Tissue hypoxia is present in many solid tumours and is linked with therapeutic resistance, aggressive tumour growth, and poor disease prognosis. Hypoxia-targeted therapies may improve patient outcome, but there is currently no clinically available method for selecting patients whose tumours contain hypoxia prior to therapy; indeed, previous clinical trials of such therapies have given poor results, likely due to a lack of determining which patients were likely to benefit. There is therefore an unmet clinical need to develop non-invasive and reliable tumour hypoxia biomarkers. Oxygen-enhanced (OE) MRI and dynamic contrast-enhanced (DCE) MRI measurements reflect tissue oxygenation and perfusion, respectively, and show potential for mapping intratumoural hypoxia. OE-MRI and DCE-MRI were performed on a cohort of 16 murine xenograft tumours (data were acquired in a separate study to this PhD). Temporal drift was observed in dynamic OE-MRI signals prior to oxygen inhalation. A model of drift-correction was developed, based on a time-varying applied flip-angle. This novel method of drift-correction was effective in characterising and correcting for signal drift across the cohort, and led to significant changes in the quantification of OE-MRI data in tumours. An optimised, data-driven analysis method (ODD) was developed to locate robustly identifiable patterns in combined OE/DCE-MRI data, and was applied to the same preclinical tumour cohort. Six tumour tissue categories were identified, two of which displayed OE-MRI and DCE-MRI enhancement characteristics consistent with hypoxia. ODD tissue segmentation agreed strongly with a separate, preclinically validated segmentation method (TBM), but without the requirement to define arbitrary data thresholds. A cohort of 12 non-small-cell lung cancer (NSCLC) patients underwent OE-MRI and DCE- MRI scans pre- and post-radiotherapy (data were acquired in a separate study to this PhD). Extensive work was carried out to process and prepare these data for subsequent combined OE/DCE-MRI analysis. ODD was applied to these new, clinical data and identified seven tissue categories, one of which likely corresponded to hypoxia. ODD tracked the response of NSCLC tumours to therapy at a microenvironmental level, and was shown to provide a more stable stratification of hypoxic vs. non-hypoxic tumours than TBM. Results from this thesis suggest: signal drift in OE-MRI data may be corrected using the methods proposed herein; ODD can map the spatially heterogeneous distribution of apparent tissue hypoxia, reliably stratify patients' tumours by hypoxic status, and monitor treatment effects; and OE-MRI and ODD are readily translatable between the preclinical and clinical research settings.

Gliomas are extremely heterogeneous, both morphologically and biologically, which contributes to a very poor prognosis. Current imaging of glioma is insufficient for a thorough diagnosis, therapy assessment and prognosis prediction. Moreover, refined and more sophisticated imaging technique could help in furthering our knowledge of gliomas. In order to facilitate proliferation, cancer cells undergo a change in structure and an increase in metabolism that results in distortion and disruption of tissue architecture. Gliomas are characterised by an increase in cells of variable sizes, as well as changes in the tissue microstructure. Diffusion-Weighted Imaging (DWI) and the apparent diffusion coefficient (ADC), have been extensively studied as potential imaging biomarkers for cellularity and tissue architecture. However, several studies have shown partial overlap in the measured values between tumour subtypes. Moreover, ADC is influenced by several factors and does not provide detailed information on the tissue microstructure. The Vascular, Extracellular and Restricted Diffusion for Cytometry in Tumours (VERDICT) is a novel diffusion model that infers tissue microstructure compartment from conventional DWI measurements. This model derives metrics for the intracellular, intravascular and extracellular– extravascular spaces providing a more detailed interpretation of the tissue microstructure. To date, VERDICT has been applied to xenograft models of colorectal cancer, patient studies of prostate cancer and recently its feasibility in glioma has been shown. In this PhD I have applied a shortened version of the VERDICT method to image intratumoral and intertumoral heterogeneity in glioma. The results have also been validated with histology as part of a prospective study. Gliomas also exhibit a significant increase in mitotic activity within the tumour. The increased number of mitosis alters cell density which, in turn, affects the total concentration of tissue sodium as the concentration of tissue sodium is approximately ten-fold higher in the extracellular compared to the intracellular space. In addition, there is a decrease in Na+/K+-ATPase activity in tumours due to ATP depletion, which contributes to disturb sodium homeostasis. Non-invasive detection of 23Na with MRI has the potential to quantify sodium concentration and therefore could be an imaging probe of cell morphology and membrane function within the tumour microenvironment, as well as a method of probing tissue heterogeneity. During my PhD, a novel 23Na-MRI technique has been used to evaluate sodium distribution within glioma and in the surrounding tissue. Metabolic reprogramming is one of the major driving forces for determining glioma growth and invasion. Therefore, the non-invasive characterization of metabolic intratumoral, peritumoral and intertumoral heterogeneity in vivo could help to better stratify patients and to develop novel therapeutic strategies targeting cancer-specific metabolic pathways. 13C magnetic resonance imaging (MRI) using dynamic nuclear polarization (DNP) is a novel technique that allows non-invasive assessment of the metabolism of hyperpolarized (HP) 13C-labelled molecules in vivo, such as the exchange of [1-13C]pyruvate to [1-13C]lactate in tumours (Warburg effect). Part of my PhD has focused on developing and translating HP [1-13C]pyruvate MRI to explore metabolic reprogramming in glioma and the surrounding microenvironment. The overall aim of my PhD has been to develop novel approaches to imaging glioma with MRI to probe both the architectural and metabolic changes of Glioma. The preliminary evidence suggests that these tools can more deeply phenotype tumours than conventional imaging approaches. Although the main focus of this work has been gliomas, the techniques developed and presented here may be applied to study other pathological conditions within the brain, which raises the possibility of other potential clinical applications for this work.

This thesis presents an investigation on multiphase reactor hydrodynamics using magnetic resonance imaging (MRI). The study demonstrates experimental techniques by which computational and quasi-analytical fluid models may be validated. Three types of industrially-important multiphase reaction vessels are considered: a co-current upflow gas-liquid-solid bed, a co-current downward trickle bed (gas, liquid, solid), and a gas-solid fluidised bed. These reactors were selected as they commonly demonstrate local hydrodynamic anisotropy which affects the global performance of industrial units. MRI was used to obtain 2D velocity images of the gas and liquid phases in the packed beds, and of the gas and the solid phases in the fluidised bed. This study reports the first spatially resolved velocity measurements of both the gas and liquid phases in a co-current upflow bed, and the gas and solid phases of an isolated bubble in a fluidised bed. The experimental vessels were: 52 mm in diameter using 5 mm glass spheres in the upflow bed at 8 bara, 27 mm with 5 mm glass spheres in the trickle bed at 6.75 bara, and 52 mm using 1.2 mm poppy seeds as the fluidised particles at 8.5 bara. The experiments were conducted at a laboratory temperature of 25.0 ± 3.0 °C. In the upflow bed, time-averaged velocity images were acquired over a 2.5 h experimental time. This was done to capture the steady state behaviour of the vessel operating in the pulsing flow regime. The temporally-stable trickle flow state in the trickle bed was imaged over 15-100 minutes. In both packed beds, severe spatial anisotropy in the distribution of flow between pores was revealed. Furthermore, the data were used to determine classical design features such as catalyst wetting and liquid holdup which compared well with literature models. The trickle bed data were further analysed using a morphological algorithm which unambiguously identified the gas-liquid and liquid-solid interfaces. The interfacial flow fields were found to be similar to the bulk flow, with most voxels exhibiting static behaviour. The amount of interaction between the phases was found to be minimal, which is typical of the low interaction regime. A single bubble injection system was employed in the fluidised bed which allowed the injection of isolated bubbles into the incipiently fluidised bed. It also enabled the triggered acquisition of NMR data at precise time intervals. The bubble was found to be an indented ellipsoidal shape, which rose with atypical behaviour which caused it to collapse. Rise velocity was found to be consistent with theory, and the injected bubbles were sufficiently spatially reproducible to acquire 2D velocity images using single-point imaging. These velocity images showed flow behaviour characteristic of a 'fast' rising bubble, with a gas recirculation cloud 37 mm in diameter. The particle field was shown to have very high flow in the bubble wake, revealing the mechanism of bubble collapse. The flow data were compared to classical two-phase fluidisation theory, which revealed noteworthy differences in the division of flow between the particulate and bubbling regions.

The environment that a child grows up in has a profound effect on their child development. For example, key outcomes such as academic ability or behaviour, cognitive ability and the neurobiology of the brain have been found to be associated to a child's environment. However, the factors that make up a child's environment are highly complex and yet the majority of research treats SES as a single number. In addition, the environment is related to several aspects of a child's development, yet there is very little research considering how these multiple levels of development relate to each other and interact. This thesis builds on the current literature by investigating how multiple aspects of a child's environment combine to create an environmental profile that is associated with positive child development. We endeavour to address three questions: 1) Which environmental factors most strongly relate to a child's academic ability, behaviour, cognitive ability and neural development? 2) Does the wider environment mediate the relationship between standard measures of SES and child development? 3) How might the environment impact academic and behaviour outcomes? In particular, is this relationship mediated by a child's cognition or the structural and functional connectivity of their brain? 7-11 year-old children (N=97) and their caregivers took part in this study. Several environmental domains and child behaviour were assessed through caregiver and child questionnaires. Academic and cognitive ability were measured using behavioural assessments. Resting state functional connectivity was measured using a magnetoencephalography (MEG) scan and structural connectivity was measured in an optional MRI scan on a separate visit (N=87). Partial Least Squares (PLS) methods identified significant relationships between the environment and child development. Multiple environmental domains were found to be reliably related to each aspect of child development. Furthermore, the wider environmental domains mediated the association between SES measures and each aspect of child development. Finally, cognition and the structural connectivity of a child's brain mediated the association between the environment and academic outcomes. This was not found for the behavioural outcomes. This thesis provides key advances towards addressing the considerable methodological challenge presented in the investigation of the complex relationships between a child's environment and multiple aspects of their development. We believe that this work will complement the available research to date and provide important detail to enable practitioners and policymakers to better support children at risk from disadvantaged environments.

Metabolism is a key process to the function and survival of all living organisms. The role of metabolism is key in many pathologies, and presents a prime opportunity to both understand and treat disease in new ways. In particular, lactate is a molecule which has been known and studied for many years by a broad spectrum of scientists from many research fields, with it appearing to have a number of roles in the routine operation and function of living organisms, as well as signaling pathological disturbance of normal physiology in many circumstances. This thesis aims to further the understanding of how measuring and quantifying both lactate and sodium with both NMR and spectral photometric techniques can be a powerful tool for the researcher in cell, pre-clinical, and clinical research. The research for my PhD has focused on basic biology, pre-clinical large animal imaging, clinical sodium imaging in Multiple Sclerosis, and a physiological first in man study of the healthy brain with hyperpolarised [1-13C] pyruvate.

The primary aim of this thesis is to investigate the transition between different flow regimes in porous media. The complete transition spectrum of single-phase flow, from creeping flow to inertial, unsteady laminar, and turbulent flow regimes, was examined in sphere packings. Further understanding of the fundamental fluid dynamics was derived based on the pore-scale flow visualisation using magnetic resonance velocimetry (MRV). Spiral imaging was selected as the ultrafast imaging protocol to probe the transient phenomena, and the acquisition was further accelerated by combining subsampling and compressed sensing reconstruction. In a random sphere packing column with column-to-diameter ratio of 3.44, the inertial effect and the onset of unsteady regime were examined with respect to the principal flow characteristics: the inertial core/channeling, backflow, and helical vortices. Helical vortices have been observed experimentally in a random packing for the first time, and the analogy between the swirling flow and helical vortices provides insight into the design and operation of packed bed reactors. Another new observation is that the transition to the unsteady regime is a highly heterogeneous process, where the evolution of the flow instability depends on the pore geometry. Moreover, pixelwise validation was achieved between the experimental and simulation results on three-dimensional velocity fields in the inertial regime; this is enabled by an image-based meshing pipeline, which reproduces the geometry of the random packing in MRV for the numerical simulation. The unsteady regimes were further investigated using a regular sphere packing, the simple cubic packing (SCP). The spectral analysis, in both the random and regular packing, revealed a route to chaos from the steady to periodic, quasi-periodic, and chaotic dynamics, which was only predicted numerically before. During the transition to turbulence, the coherent structures were extracted using proper orthogonal decomposition (POD), which yields a coherent picture regarding the turbulent dynamics, when combined with the skewness, flatness, and quadrant analysis. Furthermore, it was found that the macroscopic properties converged at lower Reynolds number than the microscopic features. In conclusion, the opportunity to measure flow fields at high spatial and temporal resolution will play an increasingly significant role in the advancement of fundamental fluid dynamics. In this thesis, MRV is used, which is particularly advantageous for non-invasive measurements in opaque systems. This thesis provides the experimental and analysis toolkits for such studies and has demonstrated the contribution to characterising and understanding different flow regimes in porous media.

Magnetic Resonance Imaging (MRI) has become a routine clinical procedure for the screening, diagnosis and treatment monitoring of various clinical conditions. Although MRI has highly desirable properties such as being completely non-ionizing and providing excellent soft tissue contrast which has resulted in its widespread usage across the gamut of clinical applications, it is limited by a slow data acquisition process. Several techniques have been developed over the years that have considerably improved the speed of MRI but there is still a clinical need to further accelerate MRI for many clinical applications. This thesis focuses on two recent advances in MRI acceleration to reduce the overall patient scan time. The first part of the thesis describes the development of a fast 3D neuroimaging methodology that has been implemented in a clinical Magnetic Resonance (MR) sequence which was accelerated using a combination of compressed sensing and sampling order optimization of acquired measurements. This methodology reduced the overall scan time by more than 60% compared to the normal scan time while also producing images of acceptable quality for clinical diagnosis. The clinical utility of accelerated neuroimaging is demonstrated by conducting a healthy volunteer study on eight subjects using this fast 3D MRI method. The results of the radiological diagnostic quality assessments that were carried out on the accelerated human brain MR images by four experienced neuroradiologists are presented. The results show that accelerated MR neuroimaging retained sufficient clinical diagnostic value for certain clinical applications. The second part of the thesis describes the development of an accelerated Cartesian sampling scheme for a rapid quantitative MR method called Magnetic Resonance Fingerprinting (MRF). This method was able to simultaneously generate quantitative multi-parametric maps such as T1, T2 and proton density (PD) maps in a very short scan duration that is clinically acceptable. The developed Cartesian sampling method using Echo Planar Imaging (EPI) is compared with conventional spiral sampling that is generally used for MR fingerprinting. The ability of novel iterative reconstruction techniques to improve the multi-parametric estimation accuracy is also demonstrated. The results show that accelerated Cartesian MR fingerprinting can be an alternative to conventional spiral MR fingerprinting.

Abdominal aortic aneurysms (AAA) are a major cause of death in men above the age of 65 in the western world. Currently decisions for AAA management are based on the size of maximum AAA diameter (>5.5cm), measured using ultrasound imaging. However, as a proportion of AAAs rupture whilst still below this diameter threshold, while larger AAAs may never rupture, better methods for AAA expansion and rupture prediction are required. Previous research suggested that the presence of "hotspots" (focal areas) of inflammation as detected with USPIO-enhanced MRI may have potential in identifying faster-growing AAAs. However, the identification of these USPIO "hotspots" had been up to this point restricted to manual processing of the MRI data in a time-consuming and laborious slice-by-slice method, which only used 2D information. Inter- and intra- observer variability were an issue, as well as the use of empirically-defined signal thresholds which were dependent on each acquisition protocol. The work presented in this thesis aimed to evaluate current methodologies for AAA assessment and growth prediction and to contribute to improved prediction models by introducing novel techniques. Ultrasound was found to under-measure AAA size and the use of maximum AAA diameter was found to be problematic, especially for growth calculations. Automatically calculated alternatives which account for the total size and shape of the AAA, as measured with MRI, were introduced for more reproducible measurements. Furthermore, automation and standardisation of the previously-employed manual methods for hotspot detection and AAA classification were achieved, with the development of an efficient algorithm with excellent agreement levels. Taken a step further, two improved algorithms were introduced, adaptive to the data and USPIO distribution of individual AAAs and eliminating the universal threshold previously used. These algorithms incorporated information on 3D USPIO distribution along the length of the AAAs to detect and visualise 3D hotspots of inflammation for the first time. Novel 2D and 3D metrics were introduced, while the algorithms were also incorporated into a GUI for ease of clinical use. Additional aneurysm metrics automatically derived by the algorithms were incorporated into multiple linear regression models to investigate prediction of AAA growth rate. This investigation introduced three significant predictors which have not been used in previous predictive models of AAA expansion: the "mean thrombus major axis" metric, which reflected baseline size of AAA throughout multiple axial slices of the AAA; the "eccentricity WT" metric which reflected the relationship between wall shape and thrombus; and the presence of "3D hotspots" which may potentially reflect transported USPIO within a network of vascular channels along the length of the aneurysm. In line with previous literature, family history of AAA and high diastolic BP were also found to be significant predictors, but larger cohorts are needed for more reliable assessment of the predictive models suggested in this thesis.

Background Aortic stenosis is the commonest valve disease requiring intervention in the developed world. Current guideline-based management strategies are based on historical observational data or expert opinion and may leave many patients with irreversible myocardial damage and adverse outcomes following valve intervention. The aims of this thesis are to investigate novel cardiac magnetic resonance techniques and how they can be applied to improve our decision making around the timing of valve intervention. Methods and Results Cardiac magnetic resonance imaging can detect two forms of myocardial fibrosis non-invasively; diffuse fibrosis using T1 mapping and replacement fibrosis with the late gadolinium enhancement technique. I devised a novel measure of diffuse fibrosis, the indexed extracellular volume (iECV) and showed that these techniques can be used to divide patients into three categories according to the type and amount of fibrosis present: no fibrosis, diffuse fibrosis and replacement fibrosis. Moreover, I demonstrated that there was evidence of increasing left ventricular decompensation across these three groups. How fibrosis and left ventricular hypertrophy change over time has not been well studied in patients with aortic stenosis. Using serial imaging scans, I showed that hypertrophy and diffuse fibrosis gradually progress over time, whilst replacement fibrosis accumulates rapidly once first established. Following valve replacement, cellular hypertrophy regresses faster than diffuse fibrosis, but replacement fibrosis appears permanent and irreversible. I then proceeded to investigate T1 mapping measures in a large international multicentre cohort of patients with aortic stenosis scheduled for valve replacement. I showed that extracellular volume-based T1 mapping measures were comparable across centres and therefore confirmed that multicentre studies are feasible. Extracellular volume fraction was associated with a decompensating ventricle and emerged as a powerful independent predictor of all-cause mortality in this group. Finally, I investigated the use of novel hybrid magnetic resonance and positron emission tomography imaging in patients with aortic stenosis, showing that this technique is feasible and well-tolerated. I tested novel attenuation correction and motion correction methods and showed that this technique can offer multiparametric imaging of valve, myocardium and coronary arteries in a single scan. Conclusion I have defined the longitudinal changes in hypertrophy and myocardial fibrosis in aortic stenosis and validated extracellular volume measures as prognostic markers in this group. Moreover, I have described novel magnetic resonance and positron emission tomography techniques and their potential to aid the clinical assessment of patients with aortic stenosis.

Dissolution dynamic nuclear polarisation is an emerging clinical technique which enables the metabolism of hyperpolarised 13C-labelled molecules to be dynamically and non- invasively imaged in tissue. The first molecule to gain clinical approval is [1-13C]pyruvate, the conversion of which to [1-13C]lactate has been shown to detect early treatment re- sponse in cancers and correlate with tumour grade. As the technique has recently been translated into humans, accurate and reliable quantitative methods are required in order to detect, analyse and compare regions of altered metabolism in patients. Furthermore, there is a requirement to understand the biological processes which govern lactate pro- duction in tumours in order to draw reliable conclusions from this data. This work begins with a comprehensive analysis of the quantitative methods which have previously been applied to hyperpolarised 13C data and compares these to some novel approaches. The most appropriate kinetic model to apply to hyperpolarised data is determined and some simple, robust quantitative metrics are identified which are suitable for clinical use. A means of automatically segmenting 5D hyperpolarised imaging data using a fuzzy Markov random field approach is presented in order to reliably identify regions of abnormal metabolic activity. The utility of the algorithm is demonstrated on both in silico and animal data. To gain insight into the processes driving lactate metabolism, a mathematical model is developed which is capable of simulating tumour growth and treatment response under a range of metabolic and tissue conditions, focusing on the interaction between tumour and stroma. Finally, hyperpolarised 13C-pyruvate imaging data from the first human subjects to be imaged in Cambridge is analysed. The ability to detect and quantify lactate production in patients is demonstrated through application of the methods derived in earlier chapters. The mathematical approaches presented in this work have the potential to inform both the analysis and interpretation of clinical hyperpolarised 13C imaging data and to aid in the clinical translation of this technique.

Metabolic imaging with hyperpolarized 13C-labeled cell substrates is a promising technique for imaging tissue metabolism in vivo. However, the transient nature of the hyperpolarization - and its depletion following excitation - limits the imaging time and the number of excitation pulses that can be used. A single-shot 3D imaging sequence has been developed and it is shown in this thesis to generate 13C MR images in tumour-bearing mice injected with hyperpolarized [1-13C]pyruvate. The pulse sequence acquires a stack-of-spirals at two spin echoes after a single excitation pulse and encodes the kz-dimension in an interleaved manner to enhance robustness to B0 inhomogeneity. Spectral-spatial pulses are used to acquire dynamic 3D images from selected hyperpolarized 13C-labeled metabolites. A nominal spatial/temporal resolution of 1.25 x 1.25 x 2.5 $mm^3$ x 2 s was achieved in tumour images of hyperpolarized [1-13C]pyruvate and [1-13C]lactate acquired in vivo. An advanced sequence is also described in this thesis in a later study to acquire higher resolution images with isotropic voxels (1.25 x 1.25 x 1.25 $mm^3$) at no cost of temporal resolution. EPI is a sequence widely used in hyperpolarized 13C MRI because images can be acquired rapidly with limited RF exposure. However, EPI suffers from Nyquist ghosting, which is normally corrected for by acquiring a reference scan. In this thesis a workflow for hyperpolarized 13C EPI is proposed that requires no reference scan and, therefore, that does not sacrifice a time point in the dynamic monitoring of tissue metabolism. To date, most of the hyperpolarized MRI on metabolism are based on 13C imaging, while 1H is a better imaging target for its 4 times higher gyromagnetic ratio and hence 16 times signal. In this thesis the world’s first dynamic 1H imaging in vivo of hyperpolarized [1-13C]lactate is presented, via a novel double-dual-spin-echo INEPT sequence that transfers the hyperpolarization from 13C to 1H, achieving a spatial resolution of 1.25 x 1.25 $mm^2$.

MRI provides a powerful tool to investigate brain structure in living humans. However a major challenge is interpreting the biological underpinnings of changes at this scale. This dissertation describes investigations into the problem of linking microscale post mortem cortical cytoarchitecture with millimeter-scale measures of cortical anatomy accessible through in vivo MRI. Chapter 1 introduces the problem and previous work done to address it. The following two chapters apply classical atlases of cortical cytoarchitecture to understanding morphological changes both in health (Chapter 2) and in disease (Chapter 3). Chapter 2 demonstrates that sensory processing hierarchies exhibit increasing gradients of cortical thickness, related to changes in cortical cytoarchitecture. In Chapter 3, cytoarchitectonically described differences in gyral and sulcal laminar structure were used to create markers of laminar change from MRI changes in schizophrenia. Classical measurements of histology have limitations; they are observer dependent, two-dimensional with limited coverage of the cortex. To address these issues, Chapters 4-6 document work carried on BigBrain, a 3D 20$\mu$m resolution histological dataset. I created a high-resolution 3D atlas of laminar cytoarchitecture, which was mapped to MRI-compatible cortical surface reconstructions. Chapter 4 records the development of an automated 1D profile-based approach to laminar analysis, revealing basic principles of cortical cytoarchitecture. In Chapter 5 this approach was extended to identify 6 cortical layers throughout the isocortex. These tools can be used to segment 1D cortical intensity profiles derived from any modality. In Chapter 6, the analysis of cortical gradients initially identified using MRI cortical thickness in Chapter 2 was replicated and extended using novel histological data. First histological cortical thicknesses were tested for the same patterns organization measured on in vivo MRI in Chapter 2. These analyses were extended to test which layers contributed most to overall thickness. High-resolution, complete maps of cortical cytoarchitecture mapped to MRI-template cortical surface reconstructions, are a powerful tool and dataset for the neuroimaging community. They offer new possibilities for linking cortical microstructure to in vivo neuroimaging.

Diffusion of water molecules in biological tissues depends on several microstructural properties. Therefore, diffusion Magnetic Resonance Imaging (dMRI) is a useful tool to infer and study microstructural brain changes in the context of human development, ageing and neuropathology. In this thesis, the state-of-the-art of advanced dMRI techniques is explored and strategies to overcome or reduce its pitfalls are developed and validated. Firstly, it is shown that PCA denoising and Gibbs artefact suppression algorithms provide an optimal compromise between increased precision of diffusion measures and the loss of tissue's diffusion non-Gaussian information. Secondly, the spatial information provided by the diffusion kurtosis imaging (DKI) technique is explored and used to resolve crossing fibres and generalize diffusion measures to cases not limited to well-aligned white matter fibres. Thirdly, as an alternative to diffusion microstructural modelling techniques such as the neurite orientation dispersion and density imaging (NODDI), it is shown that spherical deconvolution techniques can be used to characterize fibre crossing and dispersion simultaneously. Fourthly, free water volume fraction estimates provided by the free water diffusion tensor imaging (fwDTI) are shown to be useful to detect and remove voxels corrupted by cerebrospinal fluid (CSF) partial volume effects. Finally, dMRI techniques are applied to the diffusion data from the large collaborative Cambridge Centre for Ageing and Neuroscience (CamCAN) study. From these data, the inference provided by diffusion anisotropy measures on maturation and degeneration processes is shown to be biased by age-related changes of fibre organization. Inconsistencies of previous NODDI ageing studies are also revealed to be associated with the different age ranges covered. The CamCAN data is also processed using a novel non-Gaussian diffusion characterization technique which is invariant to different fibre configurations. Results show that this technique can provide indices specific to axonal water fraction which can be linked to age-related fibre density changes.

Human structural and functional brain architecture is increasingly studied by applying the mathematical framework of complex networks to data from magnetic resonance imaging. Connections (edges) in such brain networks are commonly constructed using correlations of features between pairs of brain regions, such as regional morphology (across participants) or neurophysiological time series (within participants). Subsequent analyses frequently focus on summary network statistics calculated using the strongest correlations, but often neglect potential underlying shifts within the correlation distribution. This thesis presents methods for the construction and analysis of correlation-derived structural and functional brain networks, focusing on the implications of changes within the correlation distribution. First, schizophrenia is considered as an example disease which is known to present a reduction in mean correlation between regional neurophysiological time series. Previous studies reported increased network randomisation in schizophrenia, but these results may have been driven by inclusion of a greater number of noisy edges in patients’ networks, based on retention of a fixed proportion of the strongest edges during network thresholding. Here, a novel probabilistic thresholding procedure is applied, based on the realisation that the strongest edges are not necessarily most likely to be true following adjustment of edge probabilities for effects of participant in-scanner motion. Probabilistically thresholded functional networks show decreased randomness, and increased consistency across participants. Further, applying probabilistic thresholding eliminates increased network randomisation in schizophrenia, supporting the hypothesis that previously reported group differences originated in the application of standard thresholding approaches to patient networks with decreased functional correlations. Subsequently, healthy adolescent development is studied, to help understand the frequent emergence of psychiatric disorders in this period. Importantly, both structural and functional brain networks undergo maturational shifts in correlation distribution over adolescence. Due to reliance of structural correlation networks on a group of subjects, previous studies of adolescent structural network development divided groups into discrete age-bins. Here, a novel sliding-window method is used to describe adolescent development of structural correlation networks in a continuous manner. Moreover, networks are probabilistically thresholded by retaining edges that are most consistent across bootstrapped samples of participants, leading to clearer maturational trajectories. These structural networks show non-linear trajectories of adolescent development driven by changes in association cortical areas, compatible with a developmental process of pruning combined with consolidation of surviving connections. Robustness of the results is demonstrated using extensive sensitivity analyses. Finally, adolescent developmental changes in functional network architecture are described, focusing on the characterisation of unthresholded (fully weighted) networks. The distribution of functional correlations presents a non-uniform shift over adolescence. Initially strong cortical connections to primary sensorimotor areas further strengthen into adulthood, whereas association cortical and subcortical edges undergo a subtler reorganisation of functional connectivity. Furthermore, individual subcortical regions show distinct maturational profiles. Patterning of maturation according to known functional systems is affirmed by partitioning regions developing at similar rates into maturational modules. Taken together, this thesis comprises novel methods for the characterisation of disease-related and normative developmental changes in structural and functional correlation brain networks. These methods are generalizable to a wide range of scenarios, beyond the specific disease and developmental age-ranges presented herein.

The dementias are persistent or progressive disorders affecting more than one cognitive domain that interfere with an individual’s ability to function at work or home, and represent a decline from a previous level of function. In this thesis I consider the neurophysiology of dementia at a number of levels. I investigate the ways in which the connectivity and function of the brain predisposes to the specific focal patterns of neurodegeneration seen in the various dementias. I aim to identify the mesoscopic changes that occur in individuals with neurodegeneration and how these relate to their cognitive difficulties. I show how, by assessing patients in whom there is focal disruption of brain networks and observing the outcomes in comparison to controls, I can gain insight into the mechanisms by which the normal brain makes predictions and processes language. In Chapter 1, I set the scene for the focussed experimental investigations of model diseases by beginning with an introductory, clinically-focussed review that sets out the features, aetiology, management, epidemiology and prognosis of the dementias. This places these model diseases in the context of the broader clinical challenge posed by the dementias. In Chapter 2, I turn to ‘prototypical’ model diseases that represent neurodegenerative tauopathies with predominantly cortical (Alzheimer’s disease, AD) and subcortical (Progressive Supranuclear Palsy, PSP) disease burdens. I investigate the neurophysiological causes and consequences of Tau accumulation by combining graph theoretical analyses of resting state functional MR imaging and in vivo ‘Tau’ PET imaging using the ligand AV-1451. By relating Tau distribution to the functional connectome I provide in vivo evidence consistent with ‘prion-like’ trans-neuronal spread of Tau in AD but not PSP. This provides important validation of disease modification strategies that aim to halt or slow down the progression of AD by sequestration of pathological Tau in the synapse. In contrast, I demonstrate associations consistent with regional vulnerability to Tau accumulation due to metabolic demand and a lack of trophic support in PSP but not AD. With a cross-sectional approach, using Tau burden as a surrogate marker of disease severity, I then go on to show how the changes in functional connectivity that occur as disease progresses account for the contrasting cognitive phenotypes in AD and PSP. In advancing AD, functional connectivity across the whole brain becomes increasingly random and disorganised, accounting for symptomatology across multiple cognitive domains. In advancing PSP, by contrast, disrupted cortico-subcortical and cortico-brainstem interactions meant that information transfer passed through a larger number of cortical nodes, reducing closeness centrality and eigenvector centrality, while increasing weighted degree, clustering, betweenness centrality and local efficiency. Together, this resulted in increasingly modular processing with inter-network communication taking less direct paths, accounting for the bradyphrenia characteristic of the ‘subcortical dementias’. From chapter 3 onwards, I turn to the in-depth study of a model disease called non-fluent variant Primary Progressive Aphasia (nfvPPA). This disease has a clear clinical phenotype of speech apraxia and agrammatism, associated with a focal pattern of mild atrophy in frontal lobes. Importantly, general cognition is usually well preserved until late disease. In chapter 3 itself, I relate an experiment in which patients with nfvPPA and matched controls performed a receptive language task while having their brain activity recorded with magnetoencephalography. I manipulated expectations and sensory detail to explore the role of top-down frontal contributions to predictive processes in speech perception. I demonstrate that frontal neurodegeneration led to inflexible and excessively precise predictions, and that fronto-temporal interactions play a causal role in reconciling prior predictions with degraded sensory signals. The discussion here concentrates on the insights provided by neurodegenerative disease into the normal function of the brain in processing language. Overall, I demonstrate that higher level frontal mechanisms for cognitive and behavioural flexibility make a critical functional contribution to the hierarchical generative models underlying speech perception In chapter 4, I precisely define the sequence processing and statistical learning abilities of patients with nfvPPA in comparison to patients with non-fluent aphasia due to stroke and neurological controls. I do this by exposing participants to a novel, mixed-complexity artificial grammar designed to assess processing of increasingly complex sequencing relationships, and then assessing the degree of implicit rule learning. I demonstrate that agrammatic aphasics of two different aetiologies are not disproportionately impaired on complex sequencing relationships, and that the learning of phonological and non-linguistic sequences occurs independently in health and disease. In chapter 5, I summarise the synergies between the experimental chapters, and explain how I have applied a systems identification framework to a diverse set of experimental methods, with the common goal of defining the physiology of dementia. I then return to the results of chapter 3 with a clinical focus to explain how inflexible predictions can account for subjective speech comprehension difficulties, auditory processing abnormalities and (in synthesis with chapter 4) receptive agrammatism in nfvPPA. Overall, this body of work has contributed to knowledge in several ways. It has achieved its tripartite aims by: 1) Providing in vivo evidence consistent with theoretical models of trans-neuronal Tau spread (chapter 2), and a comprehensive clinical account of the previously poorly-understood receptive symptomatology of nfvPPA (chapter 5), thus demonstrating that systems neuroscience can provide a translational bridge between the molecular biology of dementia and clinical trials of therapies and medications. In this way, I begin to disentangle the network-level causes of neurodegeneration from its consequences. 2) Providing evidence for a causal role for fronto-temporal interactions in language processing (chapter 3), and demonstrating domain separation of statistical learning between linguistic and non-linguistic sequences (chapter 4), thus demonstrating that studies of patients with neurodegenerative disease can further our understanding of normative brain function. 3) Successfully integrating neuropsychology, behavioural psychophysics, functional MRI, structural MRI, magnetoencephalography and computational modelling to provide comprehensive research training, as the platform for a future research programme in the physiology of dementia.