基于CT平扫图像的纹理分析鉴别肉芽肿性肺曲霉菌病与肺隐球菌病  

Differential Diagnosis of Granulomatous Pulmonary Aspergillosisand Pulmonary Cryptococcosis: A Non-enhanced CT Images Based Texture Analysis

作　　者：陈立鹏[1] 刘伟锋[1] 赵钊[2] CHEN Li-peng;LIU Wei-feng;ZHAO Zhao(Department of Radiology,Guangzhou First People＇s Hospital;Department of Radiology,The First Affiliated Hospital of Jinan University)

机构地区：[1]广州市第一人民医院放射科 [2]暨南大学附属第一医院影像中心,广州510180

出　　处：《中国医学计算机成像杂志》2018年第5期441-445,共5页Chinese Computed Medical Imaging

摘　　要：目的:探讨CT平扫图像纹理分析在肉芽肿性肺曲霉菌病和肺隐球菌病鉴别诊断中的应用价值。方法:回顾性分析28例表现为肉芽肿性的侵袭性肺曲霉菌病(IPA)和21例肺隐球菌病(PC)患者的肺部CT平扫图像,提取病变的纹理特征,通过Fisher、POE+ACC、Ml及三者联合(FPM)的方法选择最佳纹理参数集合,然后使用线性判别分析(LDA)和非线性判别分析(NDA)进行纹理分类。比较两种病变的纹理特征差异,进行统计学分析;并与人工鉴别的准确率进行对比。结果:纹理分析鉴别IPA和PC时,FPM联合NDA分类的误判率最低,为4.08%(2/49);PC组的纹理特征中,峰度、偏度、平方和低于IPA组,差异均有统计学差异(P<0.05);对这3个特征参数的诊断效能进行评价,绘制ROC曲线,其中峰度、偏度曲线下面积大于0.7,并获得相应的诊断截断点、敏感度及特异度;与人工鉴别相比,纹理分析具有更低的误判率。结论:肺部CT平扫图像纹理分析有助于肉芽肿性肺曲霉菌病及肺隐球菌病的鉴别诊断。Purpose： To investigate the practical value of plain CT image based texture analysis in the differential diagnosis of granulomatous pulmonary aspergillosis and pulmonary cryptococcosis. Methods： In this paper, the lung plain CT images of 28 cases of granulomatous pulmonary aspergillosis （IPA） and 21 cases of pulmonary cryptococcosis （PC） were retrospectively analyzed, and the lesion features were extracted. By Fisher, POE＋ACC, MI and the combination of the three methods （FPM）, the best set of texture parameters was selected. Then texture classification was performed with the linear discriminant analysis （LDA） and the nonlinear discriminant analysis （NDA）. The differences in texture feature between the two lesions were compared and analyzed. And the accuracy between texture classification and manual identification was also compared. Results： During the identification of invasive pulmonary aspergillosis and pulmonary cryptococcosis, FPM combined with NDA classification gave the lowest false positive rate of 4.08% （2/49）. Among the texture features of plain CT images, Kurtosis, Skewness, and sum of squares of PC group were lower than that of IPA group, and the difference was with statistical significant （P〈0.05）. The diagnostic efficiency of these feature parameters were evaluated and the ROC curves were drawn. The area under the curve of Kurtosis and Skewness was greater than 0.7, and the corresponding diagnostic cut-off point,sensitivity and specificity were obtained. Compared with manual identification, texture analysis had a lower rate of misjudgment. Conclusion： The image texture analysis of lung plain CT is feasible in the differential diagnosis of granulomatous pulmonary aspergillosis and puhnonary aspergillosis.

关 键 词：X线计算机 体层摄影术 纹理分析 侵袭性肺曲霉菌病 肺隐球菌病 鉴别诊断 

分 类 号：R445.3[医药卫生—影像医学与核医学]