This research leverages contrastive learning between MRI images and radiology reports to create a foundation model for head MRI analysis, addressing the critical shortage of annotated medical data.

• Developed SeLIP: a similarity-enhanced contrastive learning approach for multi-modal head MRI
• Utilizes existing clinical radiology reports paired with images, eliminating need for extensive manual annotations
• Significantly improves downstream tasks including segmentation and classification
• Creates more robust representations by aligning medical imagery with corresponding clinical text descriptions

This innovation matters because it makes advanced AI in medical imaging practical by reducing reliance on manually annotated datasets—a major barrier to clinical adoption of deep learning in radiology.

SeLIP: Similarity Enhanced Contrastive Language Image Pretraining for Multi-modal Head MRI