EchoNext

This study presents EchoNext, a deep learning model trained on over one million paired electrocardiogram and echocardiogram records from a large, demographically diverse health system to detect multiple types of structural heart disease. In both internal and external validation, EchoNext achieved superior diagnostic accuracy compared with expert cardiologists, maintaining strong performance across outpatient, inpatient, and emergency care settings as well as across racial and ethnic groups. In a prospective clinical trial involving patients without prior cardiac imaging, the model successfully identified previously unrecognized structural heart disease, highlighting its potential to expand early, cost-effective cardiac screening and improve diagnostic equity.