
Deep learning models applied to outpatient electrocardiograms (ECGs) in sinus rhythm successfully predict atrial fibrillation (AF) within 31 days, aiding early detection. Trained on 907,858 ECGs from six US Veterans Affairs (VA) sites and one academic medical center, the model demonstrated high accuracy (VA: 0.78, non-VA: 0.87), an area under the receiver operating characteristic curve of 0.86 (VA) and 0.93 (non-VA). The well-calibrated model, effective across diverse demographics, offers a potential tool for AF screening, reducing associated complications.
Like
Save
Share