Characterizing Alzheimer's disease molecular and anatomical imaging markers and their relationships with cognition and genetics using machine learning

Project Summary Amyloid-beta and tau are hallmarks of mild cognitive impairment (MCI)/Alzheimer’s disease (AD). The relationships of in-vivo amyloid-beta, tau, and neurodegeneration with cognitive, clinical, and genetic markers are not well understood. Patients with AD pathology exhibit heterogeneity in their clinical symptoms and illness course. Understanding the underlying neurobiological heterogeneity mechanisms of AD and improving the outcomes have been the central goals. This proposal leverages complementary information of in-vivo amyloid- beta positron emission tomography (amyloid PET), tau PET, structural magnetic resonance imaging (sMRI), cognitive, clinical, and genetic measurements via advanced machine learning methods and investigates the relationships among these measurements in patients with MCI/AD relative to normal controls. The proposal will study the data from the Alzheimer Disease Neuroimaging Initiative (ADNI; N = 898) and the Washington University’s Knight Alzheimer Disease Research Center (Knight ADRC; N = 1,121). This study will be the first to examine regional amyloid PET, tau PET, and sMRI markers and their relationships with cognitive, clinical, and genetic phenotypes using machine learning predictive modeling and heterogeneity analytics in AD research. The proposal will quantify regional PET outcomes as distribution volume ratio (DVR) and sMRI as the volumes and investigate their associations with cognitive [Mini-mental state examination (MMSE)], clinical [clinical dementia rating sum of boxes (CDR-SB) and CDR], and genetic [polygenic risk scores (PRS) and apolipoprotein E (APOE)] measurements. Aim 1 will develop machine learning modeling methods to study the relationships of amyloid PET, tau PET, and sMRI with cognitive and clinical phenotypes and test the hypothesis of whether regional brain-based imaging measurements exhibit multivariate predictive associations with cognitive and clinical phenotypes in MCI/AD patients and controls. Aim 2 will study the regional heterogeneity of amyloid PET, tau PET, and sMRI outcomes via semi-supervised machine learning methods. The study will compare the imaging outcomes between identified subgroups of patients or controls vs. each subgroup of patients to test the hypothesis of whether imaging markers differ between subgroups of patients. Aim 3 will examine the relationships of amyloid PET, tau PET, and sMRI heterogeneity signatures with cognition and genetics to test whether imaging signatures associate differentially with cognition and genetics in the subgroups of MCI/AD relative to controls. Overall, this innovative proposal will yield critical information on AD heterogeneity mechanisms, and contribute to precision medicine of diagnosis and treatment of AD. 1