PROJECT SUMMARY/ABSTRACT White matter hyperintensities (WMH) are radiologically defined regions of myelin rarefaction. WMH are found in over 90% of individuals over the age of 65, and increase the risk of other adverse outcomes, such as late-life depression, stroke, and Alzheimer’s Disease (AD). Incidence of WMH increase Alzheimer’s Disease likelihood two-fold and WMH burden correlates with AD onset, progression, and severity. Despite the broad and impactful implications of these pathologies, WMH histopathology is incompletely known due to the difficulty locating these lesions in postmortem coronal tissue. We have overcome this challenge by creating an ex vivo 7T MRI pipeline and alignment system to locate WMH in postmortem coronal slabs. Using this pipeline, we will investigate histopathologic signatures in WMH and normally appearing white matter (NAWM) to investigate theories of WMH composition previously explored in the neuroimaging literature. Neuroimaging studies demonstrate a clear cerebrovascular underpinning, but the underlying mechanism remains unknown. This project will investigate the relationship between cerebrovascular pathologies (i.e., arteriolosclerosis, perivascular space dilation, and blood brain barrier leakage) with WMH histopathologic signatures subtypes and patient outcomes. The objective of this proposed project is to conduct rigorous and robust spatial analyses of cellular and tissue architecture within WMH and NAWM tissue and investigate the correlation between these features with dementia progression and co-morbidities in Alzheimer’s Disease. Specifically, this project will evaluate pathologic correlates of various cerebrovascular abnormalities and altered cell (oligodendrocytes, microglia, macrophages, endothelial cells, pericytes), myelin, and axon distribution. Random forest regression algorithms will reveal histopathologic features with high predictive value towards WMH vs. NAWM designation. Hierarchical clustering algorithms (i.e, agglomerative nesting) will uncover clinically relevant subtypes of WMH. A weakly-supervised attention-based deep learning model will reveal regions of hematoxylin and eosin (H&E) images that are predictive of WMH vs. NAWM. We will next determine regions of WMH H&E images predictive of clinical co-morbidities (i.e., hypertension) and clinical outcomes (i.e., dementia progression) in AD patients. Together, these methods will comprehensively characterize distinct spatial relationships within WMH pathologies that correlate with patient characteristics and outcomes. These insights will inform personalized medicine approaches targeting WMH in AD and stand as a basis for future studies exploring WMH mechanisms.