Abstract
Vision-Language Models (VLMs) have demonstrated remarkable performance in medical imaging domains through contrastive learning. However, their adaptation to Magnetic Resonance Imaging (MRI) remains underdeveloped primarily due to the lack of large-scale and highquality MRI image–text datasets. To bridge this gap, we first construct a large-scale dataset comprising 2.4 million MRI image–text pairs. The dataset spans more than 18 anatomical regions and 27 disease categories, ensuring comprehensive coverage and representative diversity of the MRI domain. Building upon this dataset, we present CLMP, a model trained under the Contrastive Language-Image Pretraining (CLIP) framework enhanced with False Negative Cancellation (FNC). FNC mitigates supervision noise arising from identical textual descriptions paired with distinct images. We evaluate CLMP on linear probing and image–text retrieval tasks. Across a wide range of evaluation datasets, it achieves superior linear probing performance and strong image-text retrieval performance in median rank and Recall@1 compared with CLIP-style medical models and MR-specific baselines. Our contributions include: (1) a large-scale MRI image–text dataset, (2) a training paradigm combining CLIP and FNC, and (3) the first diagnostic-aware and naturallanguage- supervised CLIP-style model for MRI. Project page: https://goodrain553.github.io/CLMP_/.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CREATE_005.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=Xowx18QSkE
BibTex
@InProceedings{JiaHao_CLMP_MICCAISAT2026,
author = { Jiang, Haoyu AND Zhang, Chu AND Zhang, Hongyuan AND Chen, Lumin AND Wu, Zhiying AND Liu, Hongbin AND Yi, Dong},
title = { { CLMP: Contrastive Language-MRI Pretraining at Large Scale } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17275},
month = {pending},
page = {pending}
}