因子和聚类分析在继发性肺结核CT影像分类中的应用研究

目的　探讨联合应用因子分析与聚类分析统计方法在肺结核CT征象群分类的价值,为继发性肺结核病的CT分类研究提供依据。方法 将243例活动性肺结核(其中涂阳、涂阴肺结核分别为129例和114例)的肺内CT征象分为13种,应用因子分析(方差最大化正交旋转)与聚类分析相结合的方法对继发性肺结核进行影像分类.并比较在涂阳、涂阴肺结核诊断中的价值。结果 经因子分析后可形成5个公因子,即播散因子、实变因子、小叶中心正因子、小叶中心负因子、空洞因子。经聚类分析后继发性肺结核的影像类型分为5个类别,分别为空洞播散型59例(24.3%)、空洞为主型34例(14.0%)、气道播散型32例(13.2%)、实变为主型54例(22.2%)和结节型64例(26.3%)。涂阳、涂阴肺结核2组间类别构成有显著性差异(P<0.05)。其中涂阳组空洞播散型、实变型明显多于涂阴组;而涂阴组结节型明显多于涂阳组,均具有显著性差异(P<0.05)。结论 因子分析和聚类分析相结合可较好地对活动性肺结核的复杂CT征象进行科学分类,有助于活动性肺结核的诊断,特别是对涂阴肺结核的诊断具有重要价值。

Combined application of factor analysis and cluster analysis in CT images classification of secondary active pulmonary tuberculosis

Objective To explore the significance of combination of factor analysis and cluster a- nalysis in CT images classification of pulmonary tuberculosis and to provide a scientific basis for the criterion of classification.Methods The CT appearance images of 243 active pulmonary tuberculo- sis(PTB)cases were divided into 13 sorts,and analyzed by combination factor analysis(varimax) with cluster analysis.The value in diagnose smear positive and smear negative PTB was compared. Results CT findings of 243 cases can be induced into 5 common factors according to factor analysis, including air spreading factor,consolidation factor,centrilobular nodule positive factor,centrilobu- lar nodules negative factor and cavity factor.The images of the secondary PTB can be subdivided to 5 groups by K-mean cluster analysis,viz.cavity-air spread type 59 cases(24.3%),cavity type 34 cases(14.0%),air spread type 32 cases(13.2%),consolidation type 54 cases(22.2%)and centri- lobule nodule type 64 cases(26.3%).There was significant statistical difference in composition of classification between smear positive and negative cases(P<0.05).Among the total,cavity-air spread type and consolidation type in the smear positive cases were obviously more than those of negative cases,and while centrilobular nodule type in smear negative groups was vividly less than one of positive groups.Conclusions Combination between factor analysis and clustering analysis can reveal the characteristics and regularity of multiplex CT images of PTB,perform scientific clas-sification and improve the diagnosis of active PTB,especially smear negative cases.