Contrastive Joint-Embedding Prediction for Representation Learning in Structural MRI

Researchers introduced COJEPA, a self-supervised learning framework for volumetric brain MRI that combines joint-embedding predictive architecture with contrastive loss. Trained on 2,286 T1-weighted scans from two cohorts, COJEPA achieved top performance in twin retrieval, brain tumor segmentation, and age regression tasks, demonstrating effective representation learning without labeled data.

arXiv:2607.11962v1 Announce Type: new Abstract: Self-supervised learning offers a compelling approach for medical imaging, where labeled data are scarce and acquisition costs are high. We present COJEPA, a self-supervised framework for volumetric brain MRI that combines a joint-embedding predictive architecture JEPA with a contrastive loss CO , targeting two complementary properties: local predictivity and global discriminability. The model is trained without labels on T1-weighted structural MRI from two cohorts HCP-YA and AABC, $N{=}2286$, ages 22 to 90 , extending I-JEPA to 3D with foreground-aware block masking, a hierarchical convolutional patch embedding, and world-space sinusoidal positional encodings. We evaluate all three objectives across zero-shot twin retrieval, brain tumor segmentation BraTS 2024 , and age regression OpenBHB . COJEPA achieves the best monozygotic twin recall at rank@1 0.84 , the best finetuning age MAE 2.55 years on OpenBHB 3.0T , and matches CO on BraTS whole-tumor Dice, demonstrating that the combined objective yields representations that are simultaneously discriminative and locally structured.