Neuroimaging techniques are likely to continue to improve our understanding of the brain in health and disease, but studies tend to be small, based in one imaging centre and of uncertain generalisability. Multicentre imaging studies therefore have great appeal but it is not yet clear under which circumstances data from different scanners can be combined. The successful harmonisation of multiple Magnetic Resonance Imaging (MRI) machines will increase study power, flexibility and generalisability. I have conducted a detailed study of the performance of three research MRI scanners in Scotland under the name CaliBrain, with the aims of developing reliable, valid image acquisition and analysis techniques that will facilitate multicentre MRI studies in Scotland and beyond. Fourteen healthy volunteers had two brain scans on each of three 1.5T MRI research machines in Aberdeen, Edinburgh and Glasgow. The scans usually took place 2-3 weeks apart. Each scan was performed using an identical scanning protocol consisting of a detailed structural MRI (sMRI) and a range of functional MRI (fMRI) paradigms. The quality assurance (QA) of scanner performance was monitored in all three sites over the duration of the study using a three-part protocol comprising a baseline assessment, regular measures and session specific measures. The analyses have demonstrated that the data are comparable but also that within- and between-scanner variances are evident and that harmonisation work could enhance the level of agreement. The QA data suggest that scanner performance was similar between and within machines over the course of the study. For the structural MRI scans an optimised methodology was utilised to minimise variation in brain geometry between scanners and fit all the scanned brains into a common stereotactic space, such that repeated measures analyses yielded no significant differences over time for any of the three scanners. I examined the reproducibility of the fMRI motor task within and between the three sites. Similar results were obtained in all analyses; areas consistently activated by the task include the premotor, primary motor and supplementary motor areas, the striatum and the cerebellum. Reproducibility of statistical parametric maps was evaluated within and between sites comparing the activation extent and spatial agreement of maps at both the subject and the group level. The results were within the range reported by studies examining the reproducibility of similar tasks on one scanner and reproducibility was found to be comparable within and between sites, with between site comparisons often exceeding the within site measures. A components of variance analysis showed a relatively small contribution of the factor site with subject being the main source of variation. Similar results were obtained for the working memory task. The analysis of the emotional face processing task showed poor reproducibility both within and between sites. These findings suggest that multicentre structural and functional MRI studies are feasible, at least on similar machines, when a consistent protocol is followed in all participating scanning sites, a suitable fMRI task is employed and appropriate analysis methods are used.

Recently graph theory has been successfully applied to magnetic resonance imaging data. However, it remains unclear as to what the nodes and edges in a network should represent. This problem is particularly difficult when extracting morphological networks (i.e., from grey matter segmentations). Existing morphological network studies have used anatomical regions as nodes that are connected by edges when these regions covary in thickness or volume across a sample of subjects. Covariance in cortical thickness or volume has been hypothesised to be caused by anatomical connectivity, experience driven plasticity and/or mutual trophic influences. A limitation of this approach is that it requires magnetic resonance imaging (MRI) scans to be warped into a standard template. These warping processes could filter out subtle structural differences that are of most interest in, for example, clinical studies. The focus of the work in this thesis was to address these limitations by contributing a new method to extract morphological networks from individual cortices. Briefly, this method divides the cortex into small regions of interest that keep the three-dimensional structure intact, and edges are placed between any two regions that have a statistically similar grey matter structure. The method was developed in a sample of 14 healthy individuals, who were scanned at two different time points. For the first time individual grey matter networks based on intracortical similarity were studied. The topological organisation of intracortical similarities was significantly different from random topology. Additionally, the graph theoretical properties were reproducible over time supporting the robustness of the method. All network properties closely resembled those reported in other imaging studies. The second study in this thesis focussed on the question whether extracting networks from individual scans would be more sensitive than traditional methods (that use warping procedures) to subtle grey matter differences in MRI data. In order to investigate this question, the method was applied to the first round of scans from the Edinburgh High Risk study of Schizophrenia (EHRS), before any of the subjects was diagnosed with (symptoms of) the disease. Where traditional methods failed to find differences at the whole brain level between the high risk group and healthy controls, the new method did find subtle disruptions of global network topology between the groups. Finally, the diagnostic value of the networks was studied with exploratory analyses that found that, in comparison to healthy controls, people at high risk of schizophrenia showed more intracortical similarities in the left angular gyrus. Furthermore within the high risk group an increase of intracortical similarities could predict disease outcome up to 74% accuracy. The main conclusion of this thesis was that the new method provides a robust and concise statistical description of the grey matter structure in individual cortices, that is of particular importance for the study of clinical populations when structural disruptions are subtle.

No description available.

This study aims to investigate how dimensionality reduction of neuroimaging data prior to training support vector machines (SVMs) affects the classification accuracy of bipolar disorder. This study uses principal component analysis (PCA) for dimensionality reduction. An open source data set of 19 bipolar and 31 control structural magnetic resonance imaging (sMRI) samples was used, part of the UCLA Consortium for Neuropsychiatric Phenomics LA5c Study funded by the NIH Roadmap Initiative aiming to foster breakthroughs in the development of novel treatments for neuropsychiatric disorders. The images underwent smoothing, feature extraction and PCA before they were used as input to train SVMs. 3-fold cross-validation was used to tune a number of hyperparameters for linear, radial, and polynomial kernels. Experiments were done to investigate the performance of SVM models trained using 1 to 29 principal components (PCs). Several PC sets reached 100% accuracy in the final evaluation, with the minimal set being the first two principal components. Accumulated variance explained by the PCs used did not have a correlation with the performance of the model. The choice of kernel and hyperparameters is of utmost importance as the performance obtained can vary greatly. The results support previous studies that SVM can be useful in aiding the diagnosis of bipolar disorder, and that the use of PCA as a dimensionality reduction method in combination with SVM may be appropriate for the classification of neuroimaging data for illnesses not limited to bipolar disorder. Due to the limitation of a small sample size, the results call for future research using larger collaborative data sets to validate the accuracies obtained. / Syftet med denna studie är att undersöka hur dimensionalitetsreduktion av neuroradiologisk data före träning av stödvektormaskiner (SVMs) påverkar klassificeringsnoggrannhet av bipolär sjukdom. Studien använder principalkomponentanalys (PCA) för dimensionalitetsreduktion. En datauppsättning av 19 bipolära och 31 friska magnetisk resonanstomografi(MRT) bilder användes, vilka tillhör den öppna datakällan från studien UCLA Consortium for Neuropsychiatric Phenomics LA5c som finansierades av NIH Roadmap Initiative i syfte att främja genombrott i utvecklingen av nya behandlingar för neuropsykiatriska funktionsnedsättningar. Bilderna genomgick oskärpa, särdragsextrahering och PCA innan de användes som indata för att träna SVMs. Med 3-delad korsvalidering inställdes ett antal parametrar för linjära, radiala och polynomiska kärnor. Experiment gjordes för att utforska prestationen av SVM-modeller tränade med 1 till 29 principalkomponenter (PCs). Flera PC uppsättningar uppnådde 100% noggrannhet i den slutliga utvärderingen, där den minsta uppsättningen var de två första PCs. Den ackumulativa variansen över antalet PCs som användes hade inte någon korrelation med prestationen på modellen. Valet av kärna och hyperparametrar är betydande eftersom prestationen kan variera mycket. Resultatet stödjer tidigare studier att SVM kan vara användbar som stöd för diagnostisering av bipolär sjukdom och användningen av PCA som en dimensionalitetsreduktionsmetod i kombination med SVM kan vara lämplig för klassificering av neuroradiologisk data för bipolär och andra sjukdomar. På grund av begränsningen med få dataprover, kräver resultaten framtida forskning med en större datauppsättning för att validera de erhållna noggrannheten.