基于SOM聚类方法在临床检验知识发现中的应用  

Application of SOM Clustering in Knowledge Discovery of Clinical Laboratories

作　　者：孙丽君[1] 张玉海[1] 苏海霞[2] 杨显君[1] 江逊[3] 沈建军[4] 尚磊[1] 

机构地区：[1]第四军医大学军事预防医学院卫生统计学教研室,陕西西安710032 [2]第四军医大学军事预防医学院流行病学教研室,陕西西安710032 [3]第四军医大学唐都医院儿科,陕西西安710038 [4]第四军医大学唐都医院检验科,陕西西安710038

出　　处：《现代生物医学进展》2012年第33期6454-6458,6596,共6页Progress in Modern Biomedicine

基　　金：国家自然科学基金项目(30901241);陕西省科技攻关项目(2011k12-31;2009K18-01)

摘　　要：目的:探讨基于自组织特征映射网络的知识发现方法在建立应用临床检验项目的门诊患者聚类模型中的适用性,为临床检验决策支持系统的建立提供参考。方法:以2009-2011年西安市两所最大的综合医院的内科门诊患者就诊资料为训练样本,运用MatlabR2009b软件包中的SOM Tool Box工具箱建立应用临床检验项目的门诊患者的自组织映射网络聚类模型。用Excel软件绘制柱状图描述各组门诊患者的疾病特征。结果:自组织特征映射网络训练学习后,得出聚类结果,输出结果与临床检验决策真实值具有一定一致性,训练结果表明患者的聚类结果有一定的临床意义。患者性别、年龄、3年累计临床检验项目数、疾病特征对聚类模型的贡献较大。结论:自组织映射方法对应用临床检验项目的门诊患者聚类效果较好,具有较高的应用和推广价值,可为临床检验决策支持系统的研究提供方法学依据。Objective： To investigate the applicability of SOM based on knowledge discovery in the establishment of the clustering models of patients with clinical laboratory test items, and to provide reference for the establishment of decision support system of clinical laboratory. Methods： The 2009-2011 outpatients＇ data from two largest comprehensive hospital were collected. SOM tool box of MatlabR2009b software was used to build the clustering models of outpatients. Excel software was used to draw histogram to describe the disease characteristics of each group of outpatients. Results： Clustering results was got from the process of self-organizing map neural network training, the consistency between training results and real clinical laboratory was very good. The results showed that each patient cluster had similar clinical characteristics. Sex, age, disease characteristics of patients and the numbers of clinical laboratory test in three years made great contribution to the clustering models. Conclusions： Clustering effect of the SOM was satisfied, and had high value of application and popularization. This method can be used to investigate scientific evidence for the decision support system about application of clinical laboratory items.

关 键 词：临床检验 决策支持 自组织映射 聚类 知识发现 

分 类 号：R446[医药卫生—诊断学]