CT纹理分析在误诊的实性肺结节鉴别诊断中的应用  

CT texture analysis in differential diagnosis of the misdiagnosed pulmonary solid nodules

作　　者：张博薇 强金伟 叶剑定[3] 张玉 高淳[3] ZHANG Bo-wei;QIANG Jin-wei;YE Jian-ding;ZHANG Yu;GAO Chun(Shanghai Institute of Medical Imaging,Shanghai 200032,China;Department of Radiology, Jinshan Hospital,Fudan University,Shanghai,201508,China;Department of Radiology, Shanghai Chest Hospital,Shanghai Jiao Tong University,Shanghai 200030,China)

机构地区：[1]上海市影像医学研究所,上海200032 [2]复旦大学附属金山医院放射科,上海201508 [3]上海交通大学附属上海市胸科医院放射科,上海200030

出　　处：《复旦学报（医学版）》2019年第3期366-371,共6页Fudan University Journal of Medical Sciences

摘　　要：目的探讨CT纹理分析方法在误诊的肺实性结节鉴别诊断中的应用价值。方法回顾性分析CT误诊、经手术和病理证实的89例肺实性结节患者资料,包括良性病变误诊为肺癌54例和肺癌误诊为良性病变35例。采用MaZda软件对患者的CT图像进行纹理分析,分别用3种纹理特征提取方法(Fisher系数,Fisher;分类错误概率联合平均相关系数,POE+ACC;交互信息,MI)选择出前10个最有鉴别意义的纹理特征参数。采用原始数据分析(raw data analysis,RDA)、主要成分分析(principal component analysis,PCA)、线性分类分析(linear discriminant analysis,LDA)和非线性分类分析(nonlinear discriminant analysis,NDA)评估3种特征提取方法和三联法(Fisher+POE+ACC+MI,FPM)鉴别良、恶性肺实性结节的错判率(misclassified rate,MCR)。结果Fisher、 POE+ACC和MI这3种纹理特征提取方法选择的特征参数鉴别良、恶性肺实性结节的MCR均较低,FPM法可进一步降低MCR,用LDA 分析3种特征提取方法鉴别良、恶性肺结节的MCR最低;用LDA分析FPM法(LDA-FPM)可使MCR进一步降低至0。结论利用CT图像纹理分析的方法有助于对误诊的良、恶性实性肺结节进行鉴别。Objective To investigate the value of CT texture analysis in differential diagnosis of misdiagnosed pulmonary solid nodules. Methods Eighty-nine patients with solid pulmonary nodules which were misdiagnosed by preoperative CT,confirmed by surgery and pathology,were retrospectively reviewed.Among them,54 cases of benign nodules were misdiagnosed as lung cancers and 35 cases of lung cancers misdiagnosed as benign diseases.Texture analysis was performed for CT images by extracting texture features using MaZda software.The feature selection methods included Fisher coefficient (Fisher),classification error probability combined with average correlation coefficients (POE+ACC),mutual information (MI) and the combination of the above 3 methods.These methods were used to identify the 10 most significant texture features in the discrimination of benign and malignant pulmonary solid nodules. The different statistical methods including raw data analysis (RDA),principal component analysis (PCA),linear discriminant analysis (LDA) and nonlinear discriminant analysis (NDA) were used to evaluate the misclassified rate (MCR) of the 3 methods and their combination (Fisher+POE+ACC+MI,FPM)for distinguishing between benign and malignant pulmonary nodules. Results The 3 texture feature selection methods (Fisher,POE+ACC,MI)had a low MCR for distinguishing between benign and malignant pulmonary nodules.A combination of these 3 methods could further reduce MCR.Using LDA,a lowest MCR was achieved for the feature selection methods of Fisher,POE+ACC,MI,and even MCR=0 for FPM. Conclusions CT texture analysis can be used to distinguish benign and malignant pulmonary solid nodules with a good performance.

关 键 词：体层摄影术 X线计算机 肺结节 纹理分析 

分 类 号：R445[医药卫生—影像医学与核医学] R447[医药卫生—诊断学]