Detecting structural heart disease from electrocardiograms using AI.

Summary

This multicenter study develops and validates AI models that infer structural heart disease directly from 12‑lead ECGs, offering a low-cost, scalable screening approach when echocardiography access is limited. External validation across health systems supports generalizability and potential triage of patients who most need imaging.

Key Findings

• Developed deep learning models that infer structural heart disease directly from standard ECG signals.
• Demonstrated external validation across multiple health systems, supporting generalizability.
• Proposed ECG-first triage to target echocardiography for those at highest risk, potentially improving access and efficiency.

Clinical Implications

AI–ECG could prioritize echocardiography and specialty referrals, enable opportunistic screening in primary care and low-resource settings, and accelerate diagnosis of structural heart disease. Integration into EHR workflows may help surface silent disease earlier.

Limitations

• Observational model development without randomized outcome testing.
• Model performance and calibration in low-resource or community settings were not detailed in the abstract.

Future Directions

Prospective impact studies and randomized triage trials comparing AI–ECG-guided pathways versus usual care are needed; fairness assessment and calibration across demographics and devices will be critical.