基于生物信息特征提取的中医辨证.pdfVIP

第30卷第12期2007年12月 北京中医药大学学报 V01．30 No．12 Dec．20Or7 Journal of Beijing University of Traditional Chinese Medicine ·827· 基于生物信息特征提取的中医辨证 李海霞 王 阶 孙占全 何庆勇 吴 荣 张继红 西广成 胡元会 (1中国中医科学院广安门医院 北京100053；2中国科学院自动化研究所； ． 3河南省商丘市药品监督管理局) 摘要 目的 伴随着科技的发展，在多学科交叉的今天，选择合适的计算方法对中医证候进行辨证 是实现中医规范化的重要任务。方法 将多类支持向量机与基于互信息的特征提取方法相结合， 用特征提取方法多源数据进行处理，获得信息量最大的症状组合，然后利用这些症状组合作为多类 支持向量机的输入进行分类，根据支持向量机的输出就可得到相应的证候。该方法的有效性通过 一 个中医实例得到证实。结果 通过对各种方法获得的症状组合的正确辨证率对比发现，用基于 互信息的特征提取方法获得的症状组合较为理想。结论 特征提取的结果优于单纯的基于互信息 选择症状组合的结果，同时看出支持向量机适合中医证候的小样本研究，客观说明了互信息、特征 提取和支持向量机3种方法的结合适用于中医辨证分类。 关键词：中医辨证；特征提取；支持向量机；互信息 中图分类号：R241 ‘ TCM syndrome differentiation based on bioinformatics feature selection LI Hai—xia ，WANG Jie ，SUN Zhan—quan ，et a1． (1 Guang’anmen Hospital，China Academy of Chinese Medical Sciences，Beijing 100053；2 Institute of Automation，Chinese Academy of Sciences) Abstract：Objective Along with the development of sciences and technology it is the most important task for TCM standardization that the appropriate calculative method is chosen to differentiate TCM syndromes in today with the CroSS of multiple disciplines．It is the most important task in TCM that symptom group containing the most information is chosen to be differentiated。Method The feature selection based on mutual information was combined with multi—class support vector machines(SVM)， which then was treated by multiple soUrce data of feature selection for getting the symptom groups with most information．These groups were used as the inputs of multi—class SVM for classification and then the appropriate syndromes were gained according to the outputs of SVM．The efficiency of the method was approved through analyzing a TCM example