Reproduce “MR-CLIP: Efficient Metadata-Guided Learning of MRI Contrast Representations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mehmet Yigit Avci, Pedro Borges, Paul Wright, Mehmet Yigitsoy, Sebastien Ourselin, Jorge Cardoso
MR-CLIP uses metadata-guided contrastive learning to align MRI slices with DICOM acquisition descriptions, without manual contrast labels. It treats examples sharing grouped metadata labels as multiple positives. On a held-out hospital set, its 3D scan-to-text R@1 is 78.7%, versus 68.8% for ViT-B/16 InfoNCE, but InfoNCE leads on image-to-text R@5 and R@10. These results concern DICOM-derived metadata, not clinical validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.