基于加权有向图的中医量化诊断方法研究  

Study on the quantitative diagnosis of TCM based on weighted directed graph

作　　者：孙鑫亮 杨涛[2] 章颖 董海艳 胡孔法[1] 史话跃[1] 谢佳东 SUN Xin-liang;YANG Tao;ZHANG Ying;DONG Hai-yan;HU Kong-fa;SHI Hua-yue;XIE Jia-dong(Nanjing University of Chinese Medicine,Nanjing 210023,China)

机构地区：[1]南京中医药大学,南京210023 [2]南京中医药大学人工智能与信息技术学院,南京210023

出　　处：《中华中医药杂志》2020年第4期2014-2017,共4页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：国家自然科学基金项目(No.81674099,No.81503499,No.81803997,No.81804219);国家重点研发计划项目(No.2017YFC1703500);江苏高校护理学优势学科建设工程资助项目(No.2019YSHL068)。

摘　　要：目的:建立中医量化诊断方法。方法:计算医案数据中证型与症状之间的互信息,构造证型对应症状组合的联合概率分布函数,将中医诊断问题转化为有向图搜索问题,提出多标记加权有向图搜索算法(MLWDGS)。将ML-WDGS与经典的多标记学习算法(MLNB、BPMLL)在中医数据集上进行测评,以平均精度(AVP)、覆盖率(COV)、汉明损失(HL)、1-错误率(OE)、排序损失(RL)为评价指标,评价模型诊断效果。结果:ML-WDGS模型平均AVP为91.62%,COV为2.08%,HL为4.08%,OE为11.66%,RL为1.69%,总体诊断效果优于MLNB和BPMLL。结论:基于加权有向图的中医量化诊断方法能够较好的模拟中医诊断思维,诊断过程符合中医诊断逻辑,具有较好的推广和应用前景。Objective:To establish a quantitative diagnostic method of traditional Chinese medicine.Methods:The mutual information between syndromes and symptoms in medical record data was calculated,and the joint probability distribution of symptoms combination corresponding to a syndrome was calculated.The problem of TCM diagnosis was transformed into the problem of directed graph search and Multi-Label Weighted Directed Graph Search(ML-WDGS)algorithm was proposed to realize the quantitative diagnosis model of TCM.The model was compared with Multi-Label Naive Bayes(MLNB)and Multi-Label Neural Networks(BPMLL)on the data set of traditional Chinese medicine.Average precision,coverage,hamming loss,one error,ranking loss were used as evaluation indicators to evaluate the effect of the model.Results:The average precision was 91.62%,coverage was 2.08%,hamming loss was 4.08%,one error was 11.66%,ranking loss was 1.69%,the overall effect was better than MLNB and BPMLL.Conclusion:Quantitative diagnosis of TCM based on weighted digraph can better simulate TCM diagnosis.The process of model diagnosis conforms to the logic of TCM diagnosis,and has good prospects for popularization and application.

关 键 词：互信息 加权有向图 中医诊断 量化诊断 

分 类 号：R241[医药卫生—中医诊断学]