Personalized Risk Assessment for Heart Failure and Atherosclerosis in Diabetes: Radiomic Analysis of Coronary Calcium and Epicardial Adipose Tissue on Cardiac CT

Traditional cardiovascular risk assessment methods often fail to adequately distinguish between the risks of atherosclerosis-related major adverse cardiac events (MACE) and heart failure (HF) due to overlapping risk factors, limiting the effectiveness of preventive strategies. In this study we propose a novel approach to cardiovascular risk phenotyping in patients with type 2 diabetes mellitus (T2DM) by integrating advanced calcium-omics and fat-omics analyses derived from no-cost CT calcium score (CTCS) imaging. This approach allows for more precise targeting of therapeutics, such as GLP1 agonists for MACE and SGLT2 inhibitors for HF. The study cohort comprised 2662 patients with T2DM from the UH CLARIFY study. We utilized our established deep learning model, DeepFat, for segmenting epicardial adipose tissue (EAT) and engineered radiomic features of coronary artery calcifications ("calcium-omics") and EAT ("fat-omics"). Elastic net feature reduction was applied, and Cox models were trained to predict MACE and HF. Our model showed substantially higher hazard ratios (HRs) for high-risk groups, with HRs of 4.14 [2.54, 6.74] for MACE and 3.03 [1.68, 5.46] for HF, as compared to 3.88 [2.39, 6.28] and 1.88 [0.99, 3.56] with the Agatston score. Kaplan-Meier survival analyses indicated that our model captured 70.77% of MACE and 39.4% of HF cases in the high-risk groups, with a Net Reclassification Index (NRI) of 0.0335 for MACE and 0.2457 for HF, indicating more effective risk categorization compared to the Agatston score. These findings demonstrate the potential for CTCS imaging to enhance personalized cardiovascular care for patients with T2DM by facilitating targeted therapeutic interventions.