A CCTA coronary plaque radiomic model for predicting major adverse cardiovascular events in patients with coronary artery disease

Abstract

Introduction Coronary artery disease (CAD) remains the leading cause of cardiovascular mortality worldwide, and early accurate risk stratification for major adverse cardiovascular events (MACE) is critical for improving clinical outcomes. Coronary computed tomography angiography (CCTA) is the cornerstone of non-invasive plaque assessment, yet conventional morphological evaluation fails to capture plaque microheterogeneity, limiting the accuracy of traditional risk prediction. We developed and validated a multidimensional model integrating clinical, plaque morphological, and CCTA-derived radiomic features to predict MACE, and quantified the incremental prognostic value of radiomics. Methods A total of 567 patients with CCTA-confirmed CAD were retrospectively enrolled and randomly split 7:3 into training (n = 398) and test (n = 169) sets. From 1,833 extracted radiomic features, hierarchical dimensionality reduction selected core features to construct a radiomic score. Four models were established: clinical + plaque, clinical + radiomics, plaque + radiomics, and all features. Model performance was assessed using area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Incremental value was evaluated via the DeLong test, integrated discrimination improvement (IDI), and net reclassification improvement (NRI). Model interpretability was elucidated via SHapley Additive exPlanations (SHAP). Results and Discussion In the test set, the full model achieved an AUC of 0.840 (95% CI: 0.774–0.907), accuracy of 0.822, and Brier score of 0.150, significantly outperforming the clinical + plaque model (AUC = 0.771, P = 0.016). The full model yielded an IDI of 0.203 (95% CI: 0.151–0.256), categorical NRI of 0.379 (95% CI: 0.245–0.508), and continuous NRI of 1.151 (95% CI: 0.88–1.416). SHAP analysis identified texture features reflecting plaque grayscale heterogeneity as the primary drivers of MACE prediction. These findings demonstrate that combining clinical, plaque, and radiomic features significantly improves MACE prediction in CAD, and that radiomics offers substantial incremental prognostic value, providing a reliable non-invasive risk stratification tool.