Topological Image Analysis of Coronary Calcifications for Improved Cardiovascular Risk Prediction from Screening Non-Contrast CT

Accurately identifying high-risk cardiovascular patients is crucial to prevent myocardial infarctions and revascularizations. However, traditional risk assessments, such as the Agatston score, quantify coronary artery calcifications (CAC) at ≥130 Hounsfield units (HU), overlooking the contribution of lower-density calcifications. We introduce TopologyCAC, a model that leverages topological image analysis and persistent homology to capture the structure, spatial distribution, and persistence of CACs across lowered HU thresholds in CT calcium score (CTCS) images. In a cohort of 2,132 patients, TopologyCAC outperformed the Agatston score, achieving a testing Net Reclassification Improvement (NRI) of 0.11 on the testing set. Additionally, for patients under 60 years, TopologyCAC achieved a testing C-index of 0.81 in this preliminary analysis, significantly improving early-stage disease risk stratification.