Abstract 49: White Matter Hyperintensity Spatial Pattern Variations reflect distinct Cerebral Small Vessel Disease Pathologies

• Published in: Stroke (1970) Vol. 50; no. Suppl_1
• Main Authors: Phuah, Chia-Ling, Chen, Yasheng, Liu, Ziyang, Yechoor, Nirupama, Hwang, Helen, Laurido-Soto, Osvaldo, Marcus, Daniel S, Lee, Jin-Moo
• Format: Journal Article
• Language: English
• Published: 01.02.2019
• ISSN: 0039-2499; 1524-4628
• DOI: 10.1161/str.50.suppl_1.49

Abstract only Introduction: Cerebral small vessel disease (SVD) comprise of heterogeneous pathologies affecting small cerebral vessels. A novel phenotyping approach using etiology-based SVD imaging markers may improve discovery of SVD-related mechanisms. Topographical patterns seen in common SVD subtypes (hypertension-related (HTN) and cerebral amyloid angiopathy (CAA)) suggest that spatial location may reflect regionally different SVD pathology. We propose a machine-learning method leveraging spatial location information to partition WMH (a representative SVD imaging marker) into patterns that can discriminate underlying SVD pathology. Method: We selected subjects with WMH on FLAIR MRI from the Alzheimer’s Disease Neuroimaging Initiative database and created WMH segmentation maps using a custom deep learning algorithm. We performed voxel-based spectral clustering analysis on aligned WMH maps to group image voxels into clusters, maximizing within-group and minimizing between-group similarities. HTN-SVD risk was scored using serial blood pressure measurements and medication history. We quantified CAA risk using the Boston criteria. We used multivariable regression to evaluate extent to which different WMH clusters influenced CAA and HTN-SVD risks. Demographics and total WMH volume were included as covariates. Results: We analyzed 878 subjects; 39% were hypertensive, 4.4% had probable CAA. We observed partitioning of WMH into five distinct clusters, which demonstrate recognizable spatial distributions that suggest biological relevance (deep, periventricular, subcortical, anterior horns of lateral ventricles, posterior predominant). Multivariable regression models revealed independent associations of WMH clusters with CAA (posterior predominant WM; p=0.02; ß=0.63), and HTN-SVD (deep WM; p=0.01; ß=0.20). Conclusion: WMH occur in unique spatial patterns that reflect different underlying SVD pathology. WMH spatial pattern variations may serve as etiology-based SVD imaging markers, providing a critical new tool in efforts to resolve homogeneous disease mechanisms among heterogeneous disease presentations. Such strategy enhances genomic studies, allowing more comprehensive modelling of genetic effects on SVD pathogenesis.