计算机辅助形态定量法检测胃癌脱落细胞的实验

A Study on Analysis of Exfoliative Cells of Stomach Carcinoma with the Technique of Computerized Morphometry

应用计算机辅助形态定量法探讨胃癌脱落细胞的图象分析与识别。实验将胃刷涂片中的腺上皮细胞分成三类:(1)正常;(2)核异质;(3)癌。实验通过细胞图象增强,图象分割,抽取细胞的六维特征,即(1)细胞面积;(2)核面积;(3)核浆比;(4)核的形状因子;(5)核的平均灰度;(6)核内光密度的均方差。最后设计和训练分类器,从而实现了一套计算机辅助形态定量分析系统。该系统分别对样本模式、样本模式与未知样本随机混合以及未知样本重复检验,其分类符合率前二者均为98%,后者为95%。分类器的性能取得令人满意的结果。

The images of exfoliative cells of gastric carcinoma have been analyzed and recognized with the technique of computerized mophometry. The samples of cells in the stomach brush smears were divided into three classes: (1) normal, (2) dyskaryosis, (3) carcinoma. In this study, the images of cells were enhanced and segmented, Then the six-di- mentional features were extrated. They are, (1) cell area, (2) cell nucleus area, (3) ratio of cell nucleus area to cytoplasm area, (4) ratio of perimeter square to area of a cell nucleus, (5) mean gray density of cell nucleus, (6) mean square error of light density in cell nucleus, At last, the classifiers were designed and trained. Thus an interactive system of computer- ized morphometric quantitative analysis was realized. Checking the known cell samples, mixed speciments of known cell samples and unknown cell samples and unknown cell samples alone with the system, their correct classification rates are 98% in the former two groups and 95% in the latter group respectively. We consider that the function of classifiers is satis- factory.