Abstract
Accurate differentiation between infectious and non-infectious uveitis is clinically important but challenging because labeled ultra-widefield (UWF) fundus images are scarce and the two categories may share similar appearances. We propose a parameter-efficient adaptation framework based on BiomedCLIP for few-shot uveitis classification. The method first uses GPT-generated medical descriptions to construct category-level candidate knowledge and then performs visual-guided description selection to suppress irrelevant or noisy semantics. The selected medical prior is fused with learnable prompt features in the text embedding space, while an image–text contrastive objective further improves cross-modal alignment. Experiments on a UWF fundus dataset containing 438 images demonstrate that the proposed method achieves accuracies of 62.71%, 65.54%, 69.49%, and 71.19% under the 4-, 8-, 16-, and 32-shot settings, respectively, outperforming representative parameter-efficient adaptation methods in most settings. These results indicate that clinically relevant textual priors can improve the stability and discriminative ability of medical vision–language models under limited supervision.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/OMIA_034.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=hZkrD5e6jF
BibTex
@InProceedings{DiaYue_Medical_MICCAISAT2026,
author = { Diao, Yueqin AND Chen, Siming AND Wang, Yuning AND Liu, Huiying AND Xu, Yanwu AND Agrawal, Rupesh},
title = { { Medical Prior-Guided Few-Shot Adaptation of BiomedCLIP for Infectious and Non-Infectious Uveitis Classification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17270},
month = {pending},
page = {pending}
}