窗口条件对共生纹理在磨玻璃影自动识别中的影响  

The Effect of Window Settings over the Automatic Classification of Ground-Glass Opacity Based on Co-occurrence Matrix Texture Parameters

作　　者：陈旭[1] 庄天戈[1] 郑向鹏[2] 滑炎卿[2] 

机构地区：[1]上海交通大学生物医学工程系,上海200030 [2]上海华东医院CT室,上海200040

出　　处：《上海交通大学学报》2003年第z1期91-94,共4页Journal of Shanghai Jiaotong University

基　　金：上海市科技发展基金资助项目(004119011)

摘　　要：以肺部高分辨率CT(HRCT)为研究对象,探讨了常用的宽窄两种肺部CT窗口对共生纹理参数在磨玻璃影(GGO)自动识别中的影响.通过改变共生矩阵生成参数和窗口设置形成不同的初始特征集,用逐步判别法分别从中选出判别能力较强的若干特征,再由这些特征变量设计线性分类器,并用回代法和刀切法评估各分类器的性能.经比较发现,如果初始特征集只包含常用的6种共生纹理特征时,在窄窗条件下设计出的分类器对GGO的识别效果要优于宽窗条件下的;如果增加初始特征集维数,使其包含所有14个共生纹理参数时,窗口条件对分类器性能的影响可以忽略.Different feature pools were constructed by changing the combination of cooccurrence matrix formation parameters and window settings. Features that have higher discrimination abilities were selected from the respective feature pools using a stepwise discrimination method. The performance of the selected features was evaluated by training and testing a linear classifier. Once the classifier was trained, it was tested by Jackknife method as well as resubstitution. The experimental results show that: when the feature pool contains only six textural parameters that are routinely used, the performance of the classifiers designed under the wide window settings is superior to those designed under the narrow window settings. On condition that the feature pool dimension increases to 14, the effect of window settings over the performance of the classifier can be ignored.

关 键 词：计算机辅助诊断 纹理分析 共生矩阵 磨玻璃影 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] R318[自动化与计算机技术—计算机科学与技术]