Framework for the Robust Analysis of Neurological FLAIR MRI in Multi-Centre and Multi-Disease Datasets

Physicians have employed magnetic resonance imaging (MRI) of the brain to study the pathology of neurodegenerative disease. One feature that has been identified are white matter lesions (WML), which are associated with ischemic disease, multiple sclerosis, and Alzheimer's Disease. Analysis of these diseases has traditionally been conducted using T1- and T2-weighted images; however, use of the Fluid-Attenuated Inversion Recovery (FLAIR) sequence has been gaining momentum due to its superior visualization of WML. As FLAIR is relatively new, few image processing methods exist for it, and algorithms that can operate on large databases are a necessity. To address this lapse, this work presents a framework for the robust analysis of multi-centre and multi-disease FLAIR datasets. An obstacle in the development of algorithms for multi-centre datasets is the Multi-Center Effect, which encompasses variability in image properties created by differences in acquisition protocols, hardware, and artifacts. This variability can confound algorithms for automated analysis, as they often struggle to generalize to new data. Frameworks that account for this variability have not yet been developed for FLAIR MRI, and are key to analysis of big data. To account for this variability, a standardization pipeline was developed for FLAIR MRI, which normalizes image properties in large datasets. Initially, there were significant differences in the properties of images acquired by different scanners, but application of the pipeline suppressed this variability. A brain extraction algorithm was designed using a simplified model, and yielded results competitive with state-of-the-art algorithms. Ventricle segmentation and midline estimation algorithms were also developed, which, in conjunction with the other elements of the framework, permit for the extraction of several quantitative and clinically relevant metrics. These metrics were extracted from over 5,000 images volumes of FLAIR MRI, from subjects with ischemic disease, multiple sclerosis and Alzheimer’s Disease, and yielded statistically significant insights into these diseases. This thesis demonstrates that analysis using only the FLAIR modality is feasible for large studies, and presents a complete framework for analysis of multi-centre and multi-disease FLAIR MRI. This framework has the potential to revolutionize the way large-scale research will be conducted for the study of neurodegenerative disease.