Longitudinal Analysis of Brain Images for the Study of Cognitive Decline and Alzheimer's Disease Response Assessment

Treatment response assessment remains a significant challenge in the study of Alzheimer’s disease (AD). Rates of cognitive decline are heterogeneous across patients and difficult to separate from that of normal aging. It is imperative to improve the quantification of longitudinal changes in the brain, particularly considering recent breakthroughs in AD therapies. Neuroimaging is a powerful tool for quantifying brain change over time. The goal of this work was to establish quantitative, longitudinal patterns of cognitive decline in both AD and normal aging using neuroimaging. Several interpretable, increasingly complex approaches were applied including voxel-wise calculations, region-based metric extraction, Independent Component Analysis (ICA) feature tracking, and a Scaled Subprofile Modeling-based brain pattern analysis. We created extensive longitudinal imaging cohorts for development and validation using various dataset sources and explored multiple image modalities including ¹⁸F-FDG PET, T1-MRI, and ¹¹C-PiB (amyloid) PET. Performance was evaluated by applying the identified patterns to a classification task for separating cohorts of AD and normally aging subjects and implementing ROC analysis. The preliminary longitudinal brain patterns quantified in this study hold great potential for both clinical diagnostics of AD and treatment response assessment in clinical trials of AD therapies.