AI finds hidden ECG signal predicting sudden cardiac death
UC Berkeley researchers trained a model on 440,000 ECGs and identified subtle electrical patterns that predict sudden cardiac arrest far earlier than standard clinical methods, in a study appearing in Nature Medicine.
- 440,000-ECG training set; model spots pre-event signature standard screening misses
- Outperforms clinical risk scores in identifying at-risk patients before first event
- Could direct earlier ICD implant or preventive-therapy decisions in asymptomatic patients