Nat MedJClub
A clinical multimodal vision language foundation model with fine-grain explainability.
Nature Medicine · · Journal Article · Open access
Bai, Cheng + more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
EVLF-FM, a multimodal vision-language foundation model for medicine, achieves highest internal disease-diagnostic average accuracy of 0.895 across 18 global datasets spanning 10 imaging modalities.
- Why it matters: Many medical foundation models are modality-specific and provide limited transparent reasoning, which slows clinical adoption and trust.
- What they did: EVLF-FM was trained on over 1.3 million samples from 18 datasets and externally validated on 13,443 independent test samples from 17 datasets, supporting diagnosis, visual question answering, and pixel-level grounding with fine-grain explanations.
- The result: It attains Acc@0.5 of 0.842 for visual grounding across nine modalities, and human evaluation of its reasoning using S.C.O.R.E. scores averages 4.25/5, supporting clinical interpretability.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.
- Open access