基于无监督学习的心系疾病核心症状群提取研究

Research on Core Symptom Cluster Extraction of Heart Disease Based on Unsupervised Learning

目的从心系疾病临床数据中提取核心症状群。方法将聚类分析、主成分分析等无监督学习方法相结合,提出一种核心症状群提取方法。利用该方法对1741诊次心系疾病患者的临床数据进行分析。结果成功提取了3类核心症状群。“类别1”涉及14个主成分,分别对应1~5个症状,反映心气虚证和心阳虚证的主要临床特征;“类别2”涉及10个主成分,分别对应1~3个症状,反映心脾气虚证的主要临床特征;“类别3”涉及13个主成分,对应1~2个症状,反映心火亢盛证的主要临床特征。经中医专家分析,发现其能够反映心系疾病的中医证候特点。结论利用无监督学习方法可以有效提取核心症状群,从而为心系病证规律总结提供客观依据。

Objective To extract core symptom cluster from clinical data of heart diseases.Methods Combining unsupervised learning methods such as cluster analysis and principal component analysis,a new method of extracting core symptoms was proposed.1741 clinical samples with heart diseases were analyzed by this method.Results Three kinds of core symptom clusters were successfully extracted.There were 14 principal components in"category 1",relating to 1-5 symptoms separately,which reflected the main clinical characteristics of heart qi-deficiency and heart yang-deficiency syndrome.10 principal components were in"category 2",involving 1-3 symptoms respectively,which reflected the main clinical characteristics of qi-deficiency of heart and spleen.13 principal components were in"category 3",involving 1-2 symptoms respectively,that corresponded to the clinical characteristics of excessive fire in heart syndrome.After analysis by TCM experts,it was found that they could reflect the TCM syndrome characteristics of heart diseases.Conclusion The unsupervised learning method can effectively extract the core symptom group,thus providing an objective basis for the summary of the rules of heart diseases.