Biomarkers for Alzheimers disease using functional MRI data

Abstract

Investigation of novel neuroimaging related biomarkers of neurodegenerative disorders such as Alzheimer’s disease is the focus of the research work presented in this thesis. Identifying biomarkers may help to establish the gradual pathological changes in the aging brain to predict the early onset of disease. To this end, this thesis proposes an incremental framework to improve functional-magnetic-resonance-imaging (fMRI) neuroimage processing to establish biomarkers for cognitive decline and Alzheimer’s disease. fMRI data quality is poorest mostly at the cistern and cortical regions of the brain due to the environmental noise along with a much lower amplitude of the activation patterns. This results in diminishing signal-to-noise-ratio (SNR) and poses a significant challenge in the accurate analysis of the underlying blood-oxygenation-level-dependent (BOLD) changes. To counteract this, firstly a Gaussian-mixture-model (GMM) based method is developed to suppress the noise during data preprocessing. In the second stage, a method is developed to evaluate the cognitive performance of the subject by extracting the BOLD activation patterns from stimulus driven fMRI data. This method incorporated GMM along with a Bayesian framework to automate and optimize the decision making about the brain regions collaborating for a particular stimulus.Although the amplitude of the BOLD signal obtained from fMRI varies significantly among populations, there is some agreement among researchers over the response time of a working brain or the pace of the blood flow to several brain regions. Therefore in the third stage, a method is proposed to account for the BOLD fluctuations. This method, namely regional-optimum-frequency-analysis (ROFA) is based on finding the optimum regional synchrony among the accepted resting state BOLD frequency bands (e.g. 0.01Hz-0.167Hz).Furthermore, in the fourth stage a method named cluster-movement-analysis (CMA) is proposed. The CMA is based on counting the shift of individual fMRI voxels from one intensity cluster to another over the whole time-series, which has the benefit of accounting for the systematic changes in the current state of each voxel while disregarding any effect of phase similarities or differences with the neighbouring voxels. Therefore, intensity variability recorded at each voxel is used to calculate the total BOLD variability in each of the brainregions.These methods are comprehensively validated using k–fold cross-validation and hypothesis testing on synthetic as well as real fMRI data from 21 young healthy, 69 elderly healthy and 33 Alzheimer’s disease patients. Results demonstrated that the methods can significantly enhance the identification of reliable neuroimaging biomarkers for healthy ageing and Alzheimer’sdisease.

Date of Award	Apr 2016
Original language	English
Sponsors	Department of Employment and Learning (DEL)
Supervisor	Girijesh Prasad (Supervisor) & Damien Coyle (Supervisor)