UCLA

Neurological disorders are commonly characterized by disease-specific profiles of neurodegeneration that can be quantified using structural magnetic resonance imaging (MRI). Historically, most structural MRI-based studies of disease-related neurodegeneration have relied on volumetric descriptions of affected brain regions. However, reporting a single scalar summarization of a brain region’s morphometry ignores a far richer source of information contained in local descriptions of the structure’s morphometry. As a result, anatomical profiles of numerous brain disorders spanning various stages of onset, progression, recovery and their interplay with both cognitive and clinical factors remain poorly described. A promising approach to capturing local structural variations is a family of modeling techniques collectively referred to as surface-based shape analyses. In the set of studies reported in this thesis we aimed to identify shape-based biomarkers for the progression of HIV-associated neurodegeneration and prognostic biomarkers for patients with major depression likely to experience symptomatic relief following electroconvulsive therapy. Our studies’ conclusions indicate that the incorporation of shape measures are important both for descriptive and predictive modeling frameworks, beyond classical volumetric descriptors. Specifically, we report shape-based patterns of abnormal neurodegeneration among HIV+ pediatric and geriatric cohorts in regions of the basal ganglia. We further leveraged subcortical shape measures in a machine-learning framework to improve the prediction of clinical response to electroconvulsive therapy in patients suffering from major depressive disorder. Our findings suggest that including these descriptors will enhance descriptive models of neurodegeneration and may inform personalized treatment strategies.