ECG-CLIP Model Predicts Cardiovascular and Other Diseases

Eric Topol posted about a new paper in The Lancet Digital Health on ECG-CLIP, a contrastive learning foundation model trained on ECGs paired with reports. The work shows the model can predict cardiovascular diseases and outcomes, plus kidney disease, diabetes, and atrial fibrillation. Topol noted the approach uses deep learning rather than any superhuman capability. A Guardian article described the tool as spotting heart disease in under two seconds from millions of routine ECGs and potentially fast-tracking high-risk patients. The paper confirms external validation and data-efficient performance for diagnostic tasks.