AI-ECG Risk Stratification for Atrial Fibrillation: Real-World Performance and Explainability

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Artificial intelligence-enabled electrocardiography (AI-ECG) shows promise for identifying atrial fibrillation (AF) from sinus rhythm.
• The clinical utility and interpretability of AI-ECG in routine practice require further investigation.

Purpose of the Study:

• To assess the real-world performance and explainability of an AI-integrated ECG system for AF risk stratification.
• Evaluate AI-ECG's role alongside traditional clinical scores in a multicenter cohort.

Main Methods:

• 665 patients aged ≥40 years underwent 12-lead ECGs using an AI-enabled electrocardiograph (FCP-9900).
• AI assigned AF risk into four categories; machine learning models were developed and validated.
• SHapley Additive exPlanations (SHAP) assessed feature contributions.

Main Results:

• AF prevalence increased significantly across AI-ECG risk categories.
• Models incorporating clinical scores (CHADS ₂ , CHA ₂ DS ₂ -VASc) showed strong discrimination, outperforming AI-ECG alone.
• SHAP analysis identified CHA ₂ DS ₂ -VASc as the most influential predictor; AI-ECG offered modest incremental value.

Conclusions:

• AI-ECG provides rapid, low-cost AF risk estimation from a single ECG.
• Its predictive performance is modest compared to established clinical scores.
• AI-ECG may currently serve as a complementary tool, not a replacement for traditional risk stratification.