DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis

作者：Minxi Ouyang, Lianghui Zhu, Ya Bao, Qiang Huang, Jingli Ouyang, Tian Guan, Xitong Ling, Jiawen Li, Song Duan, W. Dai, Lifang Zheng, Xuemei Zhang, Yonghong He · 发表于：IEEE journal of biomedical and health informatics · 年份：2025 · DOI：10.48550/arXiv.2507.18433 · 被引用次数：1 · 研究领域：Medicine、Engineering、Computer Science

Multimodal large models have shown great potential in automating pathology image analysis. However, current multimodal models for gastrointestinal pathology are constrained by both data quality and reasoning transparency: pervasive noise and incomplete annotations in public datasets predispose vision-language models to factual hallucinations when generating diagnostic text, while the absence of explicit intermediate reasoning chains renders the outputs difficult to audit and thus less trustworthy in clinical practice. To address these issues, we construct a large-scale gastrointestinal pathology dataset containing both microscopic descriptions and diagnostic conclusions, and propose a prompt augmentation strategy that incorporates lesion classification and anatomical site information. This design guides the model to better capture image-specific features and maintain semantic consistency in generation. Furthermore, we employ a post-training pipeline that combines supervised fine-tuning with Group Relative Policy Optimization (GRPO) to improve reasoning quality and output structure. Experimental results on real-world pathology report generation tasks demonstrate that our approach significantly outperforms state-of-the-art open-source and proprietary baselines in terms of generation quality, structural completeness, and clinical relevance. Our solution outperforms state-of-the-art models with 18.7% higher clinical relevance, 32.4% improved structural completeness, and 41.2% few...