Fu-TCM model improves traditional Chinese medicine diagnosis accuracy
Fu-TCM: a multimodal large language model that learns traditional Chinese medicine diagnostic reasoning
Computation and Language
Summary
Traditional Chinese medicine (TCM) uses a detailed diagnostic process to figure out what is wrong and how to treat it. The authors built Fu-TCM, a large AI model that learns from a huge mix of text, images, and cases about TCM. This model was specially trained to follow the steps TCM doctors use to decide on a diagnosis. Fu-TCM performed better than other AI models at TCM tasks, suggesting that the reasoning behind TCM diagnosis can be learned by AI systems.
What this means in practice
- •For tcm software developers: Build AI tools that assist TCM practitioners with diagnosing syndromes by modeling traditional reasoning steps.
- •For medical ai engineers: Develop multimodal medical AI systems incorporating both text and image data for improved diagnostic reasoning.
Authors
Jiahang Xie, Tong Jin, Lingfei Wu, Yaping You, Jie Huang, Yuanchang Jin, Weikang Gong, Wei Cheng, Jia You, Pan Sun, Jianfeng Feng, Hongcheng Guo
Abstract
Large language models (LLMs) are advancing in multimodal understanding and reasoning, accelerating medical model development. Traditional Chinese medicine (TCM) uses evidence from the four examinations to determine disease location and nature, differentiate syndromes and guide treatment. This diagnostic process constitutes a complete chain of reasoning. We present Fu-TCM, a multimodal TCM LLM trained on data retained after cleaning and filtering an initial collection of approximately 7.12 million examples derived from classical texts, textbooks, medical images, clinical cases and publicly available datasets. The model was further optimized using Bianzheng-Grounded Policy Optimization (BGPO), which explicitly targets syndrome differentiation through rewards for response format, intermediate TCM reasoning fields and final syndrome predictions. Fu-TCM-27B led all six benchmarks spanning TCM reasoning, text and vision, achieving a macro-average accuracy of 81.01% and surpassing the strongest reference model with complete results across all six benchmarks by 7.13 percentage points. Its largest margin was 11.90 points on Fu-bianzheng. Case-level analyses further showed a positive association between intermediate-field accuracy and final syndrome-selection accuracy in most evaluated models. These findings suggest that TCM syndrome differentiation may be computationally learnable and that structured reasoning supervision can improve performance across TCM tasks.