AI generates heart MRI videos from cheap ECG data, boosting disease detection

A new AI model called ECGFlowCMR can generate realistic Cardiac MRI (CMR) videos from inexpensive, widely available electrocardiogram (ECG) data. This solves a major bottleneck in cardiac care, as MRIs are costly and require expert annotation. The system uses a novel framework to bridge the temporal and structural gap between the two data types. Pretraining on this AI-generated data significantly improved performance on downstream tasks like disease classification and phenotype prediction in tests on the UK Biobank dataset.