Columbia University EchoNext-Mini AI Detects Structural Heart Disease from ECGs, Outperforms 13 Cardiologists in NEJM AI Study

Columbia University researchers published the EchoNext-Mini dataset (100,000 ECG-echocardiogram pairs) and a companion AI model in NEJM AI on May 18, 2026. The model, trained on over 1 million paired ECGs and echocardiograms, simultaneously detects a broad spectrum of structural heart diseases — including valvular disease, heart failure, pulmonary hypertension, and left ventricular hypertrophy — from routine 12-lead electrocardiograms alone. In head-to-head testing, the AI outperformed 13 cardiologists. The EchoNext-Mini dataset has been released publicly, enabling other research institutions globally to build and validate similar screening tools. This is significant for healthcare equity: ECGs are available in virtually all clinical settings including rural clinics in low-income countries, while echocardiography (the current standard for structural heart disease diagnosis) requires expensive equipment and specialist operators. AI that converts widely available ECGs into echocardiography-equivalent screening could bring expert cardiac screening to hundreds of millions of patients currently without access.