MR-JEPA: A General Purpose Video Foundation Model for Cardiac MRI

iScienceLuvr · x · 2026-09-01

MR-JEPA is a self-supervised video foundation model for Cardiac MRI that extends LeJEPA to 3D spatiotemporal inputs via tubelet tokenization, spatiotemporal masking augmentation, and initialization from a 2D Cardiac MRI foundation model.

It outperforms both a natural-video foundation model (V-JEPA2) and a prior CMR-specific foundation model on all regression tasks while remaining competitive for disease classification, despite using 5x fewer pretraining videos and a smaller architecture.

Original post →

More from Research

Research channel →