MR-JEPA Video Foundation Model for Cardiac MRI

Tanishq Mathew Abraham announced MR-JEPA, a self-supervised video foundation model for Cardiac MRI. The work extends LeJEPA to 3D spatiotemporal inputs through tubelet tokenization, spatiotemporal masking augmentation, and initialization from a 2D Cardiac MRI foundation model. The linked arXiv paper states that cardiac magnetic resonance imaging produces rich sequential data such as temporal cine videos and spatial LGE/mapping stacks, yet most deep learning approaches process individual 2D slices.