人食管不典型增生上皮与原位癌的计算机纹理分析研究

采用计算机图象纹理分析和相关点阵检测技术,对人食管正常粘膜、不典型增生上皮及原位癌的不同纹理特征进行了观察。观察样品为常规病理切片,用计算机图象分析系统检测了组织的纹理特征。对受检图象建立了三种灰色分层关系矩阵,同时计算了8种纹理测度。结果显示,在重度不典型增生上皮和原位癌之间,其纹理测度和相关点阵检测数据均有显著性差异(P<0.05)。全部测量数据经计算机多元逐步判别分析,其正判率达90%以上。本研究结果表明,计算机纹理分析方法可正确地判别食管癌前病变和原位癌的组织结构异型性。提示本技术在食管癌的早期诊断方面具有肯定的实用性价值。

The study of computer texture analysis of atypical hyperplasia epithelia and carcinoma in situ of human esophagus

Using computer image texure analysis and correlation grid methods,the different texture featuresof normal mucosas,atypical hyperplasia epithelia and carcinoma in situ of human esophagus were ob-served. The samples were prepared with the routine pathologic sections and the texture features weretested in the computer image analysis system. Three kinds of the gray slicing relation matrix of the im-age were established and eight kinds of the texture measures were counted. The texture measures andthe data of correlation grid test showed significant difference between the heavy atypical hyperplasiaepithelia and the carcinoma in situ(P<0. 05). All the data were statisticaly processed by the stepwisediscriminant analysis, which got an overall accuracy of 90%. The result of the study shows that thecomputer texture analysis might distinguish correctly the structure displasia of the tissue in esophagealprecarcinomatusous change from carcinoma in situ. It also suggests that this method may have affir-mative practical value in the early diagnosis of esophageal carcinoma.