Abstract 4365988: AI-informed Coronary Artery Tortuosity Index (CArTI) from Cardiac CT Angiography Predicts 5-Year Cardiovascular Risk

Introduction: Current cardiovascular risk models rely on clinical variables and atherosclerotic burden, overlooking subtle structural variations in coronary anatomy. Coronary artery tortuosity, a marker of vascular remodeling and hemodynamic stress, may carry prognostic significance but remains underutilized due to a lack of quantifiable tools. Quantitative imaging analysis of cardaic CT angiography (CCTA) allows for extraction of high-dimensional vessel features that reflect vessel structural complexity. We developed and evaluated CArTI (Coronary Artery Tortuosity Index), a model that quantifies coronary structural variations, to assess its ability to predict 5-year risk of major adverse cardiovascular events (MACE). Hypothesis: We hypothesize that increased coronary structural complexity is associated with increased MACE risk. Methods: We analyzed CCTAs from 992 patients with MACE outcomes (median follow-up time of 4.3 years) determined via ICD and CPT codes. We segmented coronary arteries using a 3D U-Net. From each segmentation, we extracted 227 features quantifying vessel structure, such as tortuosity, curvature, and torsion. We split the dataset 50/50 into training and testing (each n=496) cohorts and applied univariate filtering across 50 bootstrap iterations to identify the most predictive and stable features. Using the top 8 features, we trained a Cox proportional hazards model to assess MACE risk over a 5-year period. We stratified patients into high- and low-risk groups based on median predicted risk and evaluated model performance using concordance index (C-index), hazard ratios (HR) and Kaplan-Meier (KM) curves. Results: On the holdout test set, CArTI successfully stratified patients’ 5-year MACE risk with a C-index of 0.648. KM analysis demonstrated separation in survival between high- and low-risk groups with high-risk patients showing an HR of 2.69 (95% CI: 1.07–6.78, p<0.05) compared to low-risk patients. Among the top predictive features that were positively associated with MACE risk were measures of curvature and tortuosity. Conclusion: CArTI, a quantitative model of coronary artery structural complexity derived from CCTA, predicts 5-year MACE risk and stratifies patients into clinically meaningful high- and low-risk groups. These findings strongly suggest that vessel complexity encodes prognostic information and highlights the potential of explainable AI tools to advance non-invasive MACE risk assessment. Further validation is warranted.