Arrhythmias & electrophysiology
Image based artificial intelligence-enhanced electrocardiogram prediction of incident atrial fibrillation
Authors' conclusion
This image-based approach enables AI-ECG prediction of AF in settings without digital ECG infrastructure and provides additive value to known clinical risk scores.
Study context
- Population
- Patients with ECGs in the Beth Israel Deaconess Medical Center (BIDMC) dataset and participants in the United Kingdom (UK) Biobank189,539 BIDMC + 65,610 UK Biobank
- Comparison
- Image-based AI-ECG vs CHARGE-AF risk score
- Outcome
- Prediction of incident atrial fibrillation (AF) (C-statistic)
Methodological detailsmulticentre, external validation, Saliency mapping: model focused mainly on P-wave morphology and PR interval regions
- Multicentre
- External validation
- Saliency mapping: model focused mainly on P-wave morphology and PR interval regions
Performance
As reported in the abstract
Incident AF prediction (C-statistic)
C-statistic, BIDMC dataset
0.754
95% CI 0.747–0.761
C-statistic, UK Biobank
0.723
95% CI 0.704–0.741
Further findings
Performance in subgroups (outpatients, women, non-white individuals)
- Performance maintained across key subgroups
AI-ECG vs CHARGE-AF risk score
- C-statistic, AI-ECG model vs CHARGE-AF (superior)
- 0.696 vs 0.667, P < .05
- C-statistic, combined (significant additive value)
- 0.711, P < .0001
Smartphone-photographed ECGs
- C-statistic
- 0.736
Saliency mapping
- Model primarily focused on P-wave morphology and PR interval regions
Original publication
Image based artificial intelligence-enhanced electrocardiogram prediction of incident atrial fibrillation. Heart Rhythm. Published 14 October 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →