Conceptual Framework for Multimodal Data Fusion in Precision CAD Risk Prediction.

Pivotal Methodological Shift in Biomedical Informatics for Cardiovascular Health. Ongoing Informatics Challenges and Future Research Directions Emphasized.

Heterogeneous patient data sources—including imaging biomarkers (e.g. CAC, CCTA), genomic/PRS information, longitudinal EHR trajectories, and wearable-device/sensor signals—feed into an AI-enabled fusion engine that combines automated feature extraction with model architectures (e.g. CNNs, RNNs, GNNs) and explicit fusion strategies (early/intermediate/late), while addressing missingness, heterogeneity, and interpretability (XAI). The resulting models aim to improve discrimination and reclassification and to enable individualized, actionable risk assessment. Study-level performance metrics (AUC/C-index, calibration, and reclassification indices) are summarized in Table 1 and Supplementary Table S1.

CAC, coronary artery calcium; CCTA, coronary CT angiography; CNNs, convolutional neural networks; EHR, electronic health record; GNNs, graph neural networks; PRS, polygenic risk score; RNNs, recurrent neural networks; XAI, explainable AI.