Analyzing Mild Cognitive Impairment Progression via Multi-view Structural Learning

Abstract

Alzheimer’s disease (AD), and its precursor, mild cognitive impairment (MCI), are progressive neurodegenerative conditions with a preclinical period that can last a decade or more. A variety of predictive models and algorithms have been developed to classify different clinical groups (e.g., elderly normal versus MCI) or predict conversion (e.g., from MCI to AD) based on longitudinal neuroimaging or other biomarker datasets. Even so, it is still unknown how brain structural and functional alterations jointly contribute to the MCI/AD progression process. Here we introduce a novel supervised multi-view structure learning framework to model the latent patterns of MCI/AD progression. Specifically, we learned and optimized a common data representation based on both structural and functional connectome data. Instead of determining patterns of abnormal structural and functional connectivity and their overlap, we create and analyze a common structure (graph) that can describe the entire process of disease progression. Different structural and functional connectome features contribute to this common structure simultaneously. The learned common structure reflects “a progression path” of MCI: it starts from elderly normal, proceeds through significant memory concern (SMC), early MCI and eventually ends with late MCI. As the common structure is learned from different structural and functional connectome features, it suggests that the connectome alterations related to MCI progression might happen in different structural and functional regions simultaneously.