Artificial Intelligence Applied to Electrocardiograms Recorded in Sinus Rhythm for Detection and Prediction of Atrial Fibrillation: A Scoping Review

Summary

This comprehensive scoping review synthesises 19 studies applying AI to ECG recordings in sinus rhythm for detecting or predicting paroxysmal or new-onset atrial fibrillation (AFib). The key clinical motivation is that AFib is often paroxysmal (intermittent) and may not be present during a clinical ECG recording, yet DL models can detect subclinical atrial changes from sinus rhythm ECG morphology that predict future AFib onset. The review covers the Mayo Clinic lineage (AUROC 0.87 for AF detection from sinus rhythm), Mass General Brigham, UK Biobank, and multinational hospital cohorts, with training datasets ranging from a few thousand to over 1 million ECGs.