Frontiers in Cardiovascular Medicine · Oct 30, 2026
Artificial Intelligence–Enabled Multimodal Predictive Models for Atrial Fibrillation Risk Stratification: Current State and Future Directions
By Shishi Huang, Zhenyi Li, Weiqiang Wang et al.
Atrial fibrillation (AF) remains a major contributor to the global stroke burden and serious cardiovascular outcomes such as heart failure. Traditional single-modality assessments have limited discriminatory power for reliable prediction of new-onset AF. This review synthesizes recent advances in AI-driven multimodal approaches for incident AF prediction, encompassing electrocardiography, cardiac imaging, biomarkers, and electronic health records. While conventional clinical risk scores (e.g., CHARGE-AF, C2HEST) provide baseline stratification, and emerging single-modality AI shows promise, current multimodal evidence remains heterogeneous and largely exploratory. Fused models may offer incremental value over single-modality or clinical approaches in selected cohorts; however, direct comparison of AUC or NRI across studies is limited by heterogeneity in endpoints, follow-up, and validation design. Key challenges include data heterogeneity, model interpretability (XAI), external validation, and clinical implementation. Future efforts should prioritize prospective evaluation, cross-modal attention mechanisms, and equitable deployment to translate multimodal AI into precision AF prevention.