AI-driven radiomics and functional imaging integration for intelligent identification of coronary high-risk plaques and prediction of cardiovascular events.

BACKGROUND

Accurate identification of high-risk coronary plaques and prediction of cardiovascular events are essential for improving the management of coronary artery disease (CAD). Although coronary computed tomography angiography (CTA) is widely used in clinical practice, conventional assessment methods remain limited by subjective interpretation and suboptimal prognostic accuracy. This study aimed to develop and validate a deep learning (DL)-based radiomics model integrating coronary CTA and computed tomography-derived fractional flow reserve (CT-FFR) for automated identification of high-risk coronary plaques and prediction of cardiovascular events.

METHODS

In this retrospective study, 800 coronary CTA scans from patients with stable plaques, vulnerable plaques, and acute coronary syndrome (ACS) were analyzed. Clinical characteristics and follow-up outcome labels were incorporated for annotation of major adverse cardiovascular events (MACE) during follow-up. After image preprocessing, coronary arteries were automatically segmented using a U-Net-based DL model. Radiomics features describing plaque morphology, texture, and attenuation, including surrogates for fibrous cap integrity, were extracted and combined with CT-FFR to incorporate functional hemodynamic information. A multimodal fusion model integrating convolutional neural networks (CNNs) and Transformer architecture was developed and internally validated using cross-validation.

RESULTS

In the independent test cohort, the multimodal model achieved an accuracy of 90.8%, sensitivity of 91.2%, specificity of 89.4%, F1-score of 0.91, and area under the curve (AUC) of 0.927 [95% confidence interval (CI): 0.902-0.948], outperforming the structural-only model (AUC =0.864) and functional-only model (AUC =0.878). Feature importance analysis indicated that the radiomics-derived fibrous cap surrogate, plaque attenuation, and CT-FFR were among the most influential predictors of adverse cardiovascular outcomes.

CONCLUSIONS

This study presents an interpretable multimodal framework for quantitative assessment of high-risk coronary plaques and cardiovascular risk prediction using CTA. The proposed approach improves risk stratification and supports the integration of artificial intelligence (AI) into clinical management of CAD.