Dual-Phase Cross-Modal Contrastive Learning for CMR-Guided ECG Representations for Cardiovascular Disease Assessment

Researchers have developed a new AI model that learns to predict detailed heart structure and function from simple, cheap electrocardiograms (ECGs) by training on over 34,000 paired ECG and 3D cardiac MRI scans. The dual-phase contrastive learning framework aligns ECG data with 3D heart anatomy, improving the extraction of functional parameters by 9.2%. This could enable scalable, cost-effective cardiac screening by deriving MRI-like insights from ubiquitous ECGs.