Abstract P710: Using Dictionary Learning to Determine Differences in Vasculature Between Males and Females to Understand Gender Differences in Stroke

Abstract

Introduction: Between males and females, there is a variation in stroke incidence, with males having a stroke incidence 33% higher than women. This difference may be partly due to gender differences in vascular regions between males and females. We hypothesize that statistically significant population differences in vasculature may be identified and measured through image analysis methods. Methods: We have developed and verified a scalable big data image-processing pipeline for magnetic resonance angiography (MRA), without relying on a MRI registration intermediary, and allowing for multi-modality analysis. Our approach uses dictionary learning to find differences between groups by comparing projections of 3D images from each individual. It enables the degree of regional variation between population groups to be visualized as a color map overlaid on a representative template of the Circle of Willis (CoW). Results: Analyzing a dataset of MRA images from 42 patients (25 female, 17 male), we found the group of females to have larger ICA and more likely to have a complete CoW. The figure presents a heat map of differences between male and female patients from our analysis. Conclusion: The gender differences we found with this new method are consistent with previous research. This multi-modality population level analysis tool may be used to analyze stroke imaging data and extract key global differences. With big data available today, it can be used to quantify statistical differences in vasculature between populations, to gain insight for better stroke diagnosis or treatment.