病理图像分析中的两步骤聚类分析应用  

Application of two step clustering in pathological image analysis

作　　者：关鹏[1] 全宇[2] 何苗[3] 周宝森[1] 

机构地区：[1]中国医科大学公共卫生学院流行病学教研室,沈阳110001 [2]中国医科大学附属第二医院计算机室 [3]中国医科大学附属第一医院计算机室

出　　处：《中国公共卫生》2006年第10期1264-1265,共2页Chinese Journal of Public Health

基　　金：辽宁省教育厅高等学校科学研究项目(202013137;05L534)

摘　　要：目的探讨两步骤聚类分析及其在病理图像诊断分析中的应用。方法对正常、低度鳞状上皮内病变和高度鳞状上皮内病变宫颈细胞的51个特征参数采用两步骤聚类分析:(1)将样品预聚类成小的子类;(2)对预聚类的子类再进行逐步聚类。聚类采用对数似然距离,根据贝叶斯信息准则自动决定适宜分类数目,并对各指标重要性进行度量。结果对于正常、低度鳞状上皮内病变和高度鳞状上皮内病变宫颈细胞的分类正确率分别为98.0%,96.1%和100%。结论该聚类分析方法分类正确率较高,分类中各指标重要性的度量对指导病理图像分析具有一定的实际意义和应用价值。Objective To investigate the potential of two step clustering analysis in pathological image analysis. Methods Two step clustering was used to cluster normal, low - grade squamous intraepithelial lesion, high - grade squamous intraepithelial lesion cervical cells with 51 characters. Firstly, the cases were clustered intomany small sub - clusters and then the sub -clusters were clustered into the desired number of dusters. The number of clusters was automatically determined by Bayesian Information Criteria with log - likelihood distance measure. Results Classification accuracy of normal, low - grade squamous intraepithelial lesion and highgrade squamous intraepithelial lesion cervical cells was 98.0 %, 96.1% and 100 %, respectively. Conclusion With the high classification accuracy, the two step clustering can measure the variables＇ significance, which could provide useful information for image analysis.

关 键 词：两步骤聚类 图像分析 宫颈细胞 

分 类 号：R36[医药卫生—病理学]