AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
PHOENIX AI: A Multimodal Foundation Model for Coronary Plaque Phenotyping and Prediction of Future Acute Coronary Syndrome Using Coronary CT Angiography, A Global Multicenter Retrospective Study
By Singh S, Reddy S, pillay S et al.
Introduction: Despite advances in preventive cardiology, acute coronary syndrome frequently occurs in individuals without prior cardiovascular events and remains inadequately predicted by conventional clinical risk factors. Current risk assessment models fail to capture underlying plaque vulnerability. Accurate identification of high risk coronary plaque phenotypes before clinical rupture remains a major unmet need in cardiovascular medicine. Methods: We conducted a global multicenter retrospective study of 60,000 patients undergoing coronary CT angiography across 12 international centers. PHOENIX AI, a multimodal foundation model, integrated automated deep learning-derived plaque phenotypes, clinical variables, and inflammatory biomarkers. The model was developed in 45,012 patients and externally validated in 14,988 patients. Results: Among 60,000 patients (mean age 62.0±10.5 years; 58.0% male), 3,026 incident acute coronary syndrome events occurred during 26.8±14.5 months of follow-up. Low attenuation plaque burden independently predicted future ACS (OR 1.16 per 1% increase; 95% CI 1.15-1.17; p<0.001), as did positive remodeling (OR 9.29; 95% CI 6.38-13.53; p<0.001), napkin ring sign (OR 3.35; 95% CI 3.05-3.67; p<0.001), hs-CRP (OR 1.06; 95% CI 1.03-1.09; p<0.001), and age (OR 1.02; 95% CI 1.02-1.03; p<0.001). ACS rates increased from 2.19% (95% CI 2.00-2.38) in patients without high-risk plaque features to 4.11% (95% CI 3.86-4.36), 10.46% (95% CI 9.89-11.03), and 20.29% (95% CI 18.40-22.18) in those with one, two, and three features, respectively (p-trend<0.001). In external validation, PHOENIX AI significantly outperformed the clinical model (AUROC 0.741, 95% CI 0.723-0.759 vs 0.589, 95% CI 0.567-0.611), yielding a ΔAUROC of 0.152 (95% CI 0.123-0.181). Conclusion: PHOENIX AI accurately identified vulnerable coronary plaque phenotypes and substantially improved prediction of future acute coronary syndrome beyond traditional clinical risk factors. The strong stepwise association between high-risk plaque burden and adverse outcomes, coupled with robust external validation, supports its potential role in personalized cardiovascular risk stratification and prevention.