P4‐101: BRAIN AGING HETEROGENEITY ELUCIDATED VIA MACHINE LEARNING: THE MULTI‐SITE ISTAGING DIMENSIONAL NEUROIMAGING REFERENCE SYSTEM

Abstract

Brain aging is affected by a number of underlying neuropathological processes, which are typically manifested by different neuroanatomical, functional and neuropathological patterns. Disentangling this heterogeneity and associating it with demographic, cognitive, clinical and genetic risk factors for AD is challenging, and requires sophisticated multi-variate methodology and large sample sizes. Neuroimaging offers complex multi-parametric information on patterns of brain structure, functional connectivity, as well as amyloid and tau burden. We formed the iSTAGING consortium, a large-scale pooling of brain scans from over 15,000 subjects/16 studies/dozens of sites (Fig. 1), aiming to understand and visualize brain aging patterns and relate them to non-imaging measures. Preliminary results from six cohorts (ADNI1, ADNI2, SHIP, BLSA1, BLSA2, Penn ADC; n=3,179) covering a heterogeneous spectrum of cognitively normal brain aging, MCI and AD were harmonized and used to prototype a dimensional system. For clarity, three dimensions are visualized herein, using machine-learning: 1) SPARE-BA, a subject-specific index of age-related brain atrophy patterns; 2) SPARE-AD, an index previously developed to capture AD-like atrophy patterns; 3) white matter hyperintensities (WMH) burden. Individual brain regions were also measured but are not displayed herein. The cross-sectional variation according to clinical diagnosis (CD) and APOE4 genotype was visualized in this 3D prototype iSTAGING refe…