IC‐P‐084: Interaction of cerebrovascular lesions and atrophy on brain MRI in relation to cognitive decline over four years in people with symptomatic atherosclerotic disease: The SMART‐MR study

Abstract

Background: While volumetric assessments of the brain can provide insights into atrophy for specific regions of interests, anatomical shape patterns characterizing both regional abnormalities and their precise localizations are often overlooked. Through analysis of the shapes of subcortical brain structures, this study aims to investigate the patterns of brain atrophy between a highly selective North American cohort and a community-based Asian cohort. Datasets used were from the Alzheimer Disease Neuroimaging Initiative (ADNI) and the Memory, Aging and Cognition Centre (MACC) study respectively, both large studies which have acquired structural MRI measures for the identification of biomarkers for dementia. We hypothesize that imaging markers derived from ADNI are applicable to the MACC population, regardless of their differences in vascular risk factors and cerebrovascular disease.Methods: Out of the 299 subjects analyzed, 189 were from ADNI (AD1⁄456,NC1⁄4133) and 110 were from MACC (AD1⁄423, NC1⁄487). Shape changes of subcortical brain structures (hippocampus, amygdala, caudate, globus pallidus, putamen, thalamus) and lateral ventricles were characterized using computational anatomic mapping methods (Large Deformation Diffeomorphic Metric Mapping (LDDMM)-surface mapping algorithm). Data reduction (Multi-dimensional scaling) was employed to produce low dimensional embeddings of the shape signatures, which were in turn used as features in a soft-margin support vector machine (SVM). An SVMmodel was first created using the ADNI cohort before using theMACC cohort as a replication dataset for validation. Results: Both ADNI and MACC cohorts showed similar patterns in brain atrophy, with classification rates of agreement varying from 70-80% between NC and AD. When the classifier trained from ADNI was applied to MACC, comparable sensitivities and specificities were observed in the hippocampus shape (classification rate 75%, specificity 75%, sensitivity 74%), amygdale shape (classification rate 76%, specificity 76%, sensitivity 78%) and ventricles shape (classification rate 77%, specificity 78%, sensitivity 74%). Results were improved when the shape signatures from individual structures were combined, resulting in higher classification rates (hippocampi and ventricles: specificity 82%, sensitivity 74%; hippocampi, ventricles, amygdala: specificity 79%, sensitivity 74%). Conclusions: Despite vascular risk differences between the ADNI and MACC cohorts, structural imaging markers extracted from the ADNI cohort are found to be applicable to a community-based Asian cohort.