目的:本研究旨在利用中医临床医案数据和知识蒸馏技术,构建具备推理能力强、可信度高的中医辨证论治智能诊疗模型。方法:以GPT4o为教师模型,对中医医案数据进行知识蒸馏,生成高质量的中医辨证论治指令数据集,并基于Qwen2.5-7b模型采用LoRA方法进行监督微调,以增强其中医诊疗推理能力和个体化辨证论治能力。结果:本研究提出的知识蒸馏微调方法大幅提升了中医诊疗推理过程的透明性和可解释性,并保留处方推荐的精准性,表明模型生成的文本可读性更高,诊疗推理能力更强。结论:采用知识蒸馏的中医辨证论治大模型能有效提升诊疗推理和个体化辨证论治能力,为中医智能化诊疗和临床辅助决策提供了新思路。

Objective:This study aims to develop an intelligent diagnosis model for Traditional Chinese Medicine (TCM) syndrome differentiation and treatment with strong reasoning capabilities and high reliability, using clinical case data and knowledge distillation techniques.Methods:The GPT4o was employed as the teacher model to perform knowledge distillation on TCM clinical case data, thereby generating a high-quality TCM syndrome differentiation instruction dataset. Subsequently, the Qwen2.5-7b model was fine-tuned using the LoRA method under supervised learning to enhance its reasoning ability in TCM diagnosis and individualized syndrome differentiation.Results:The proposed knowledge distillation fine-tuning approach significantly improved the transparency and interpretability of the TCM diagnostic reasoning process, while maintaining the precision of prescription recommendations. The results indicate that the generated text has higher readability and that the model exhibits stronger diagnostic reasoning capabilities.Conclusion:The TCM model enhanced by knowledge distillation effectively improves diagnostic reasoning and individualized syndrome differentiation, offering a novel approach for intelligent TCM diagnosis and clinical decision support.

国家自然科学基金委员会面上项目(82074580):基于知识图谱的现代名老中医诊治肺癌用药规律及其机制研究,负责人:胡孔法；国家自然科学基金委员会面上项目(82174276):知识和数据协同驱动的中医藏象智能辨证方法研究——以心系疾病为例,负责人:杨涛；江苏省中医流派研究院开放课题(JSZYLP2024060):基于大模型技术的江苏中医流派知识挖掘和服务创新方法学研究 ,负责人:杨涛。