摘要
目的:利用改良Transformer算法构建冠心病中医证候诊断、方药推荐模型。方法:以冠心病证候要素为关键环节,基于临证诊疗思路“症状-证候要素-证候-治法-方剂-药物(剂量)”搭建基本逻辑,综合运用多头注意力机制、复合词向量、随机失活形成改良Transformer算法,模拟临床医师临证思路,形成具备冠心病中医证候要素判断、证候诊断、方药推荐、可更新迭代功能的智能化模型。模型建立后,选择8030例临床病例诊疗数据作为训练集进行模型训练,随机筛选100例基于真实临床病例的中医开方数据进行测试,比较模型输出方药与临床医师方药,对模型进行定性评价。结果:加入多头注意力机制的改良Transformer算法使模型准确率有更大的提升,模型在主要证候的判断、主要方剂的选择上与临床医师一致,在药物加减上有一定提升空间。结论:改良Transformer模型可提高冠心病证候要素、主要证候、处方、用药的准确率,较为准确、稳定地输出主要证候和方药建议,是中医智能化发展的体现。
Objective:To construct a traditional Chinese medicine(TCM)syndrome diagnosis and prescription model for coronary heart disease with the improved Transformer algorithm.Method:Taking the syndrome elements of coronary heart disease as key links,the model was constructed based on the clinical diagnosis and treatment principle of"symptoms-syndrome elements-syndrome-treatment method-prescriptionmedicine(dose)".The basic logic of improved Transformer algorithm was constructed with multi-head attention mechanism,compound term vector and dropout,in order to form the model with functions of TCM syndrome elements judgment,syndrome diagnosis,prescription recommendation.After the model was constructed,it was trained by 8030 cases.And 100 cases with TCM prescriptions were randomly selected for testing,and the model output prescriptions were compared with those of clinicians for qualitative evaluation of the model.Result:The improved Transformer with multi-head attention improved the accuracy of the model.The model was consistent with clinicians in the judgment of main syndromes and the selection of prescriptions.Whereas there was a certain room for improvement in the addition and subtraction of medicines.Conclusion:The improved Transformer model can improve the accuracy and stability of output,which is an embodiment of the intelligent development of TCM.
作者
李洪峥
王阶
郭雨晨
张振鹏
李剑楠
李谦一
董艳
杜强
LI Hongzheng;WANG Jie;GUO Yuchen;ZHANG Zhenpeng;LI Jiannan;LI Qianyi;DONG Yan;DU Qiang(Guang′anmen Hospital,China Academy of Chinese Medical Sciences,Beijing 100053,China;School of Information Science and Technology,Tsinghua University,Beijing 100084,China;Shenzhen International Graduate School,Tsinghua University,Shenzhen 518055,China;Imperial College London,London SW72AZ,United Kingdom)
出处
《中国实验方剂学杂志》
CAS
CSCD
北大核心
2023年第1期148-154,共7页
Chinese Journal of Experimental Traditional Medical Formulae
基金
国家中医药管理局中医传承与创新“百千万”人才工程(岐黄工程)——国家中医药领军人才支撑计划岐黄学者项目(0201000401)
国家中医药管理局冠心病中医人工智能循证能力建设项目(60103)。
