Predicting major adverse cardiovascular events in diabetic and non-diabetic patients with coronary artery disease: visual models integrating multi-parametric coronary computed tomography angiography and pericoronary adipose tissue radiomics.

Objective: To compare the application value differences of PCAT radiomic features, clinical risk features and computed tomography (CT)-derived parameters in predicting Major adverse cardiovascular events (MACE) in patients with/without diabetes. Methods: Retrospective analysis included 1,000 coronary atherosclerosis patients undergoing Coronary CT angiography (CCTA) (with/without diabetes: 274/726) from the Eighth Affiliated Hospital of Southern Medical University. Clinical/CT data were collected, extracting 285 PCAT radiomic features from three major coronaries. Least absolute shrinkage and selection operator regression identified MACE-associated radiomic features. Patients underwent random 6:4 training/testing cohort split. Four predictive models were constructed: Model 1 (clinical factors), Model 2 (imaging factors), Model 3 (imaging-radiomic features), Model 4 (all factors). Results: In the training set, Model 4 showed the best performance: The area under the curves (AUC) of 0.803 [95% confidence interval (CI): 0.756-0.850] and 0.854 (95% CI: 0.779-0.929) for groups with/without diabetes, respectively. Model 3 outperformed Model 2 in patients without diabetes ( Conclusion: PCAT radiomics, CT-derived parameters, and plaque features demonstrate differential predictive value for MACE in patients with/without diabetes. Combining these with clinical risk factors provides most effective model for both.