基于常规磁共振成像的图像纹理分析对肝局灶性病变的鉴别诊断价值  

Texture-based classification of focal liver lesions on conventional magnetic resonance imaging

作　　者：夏鹏[1] 王国祥[1] 刘栋[1] 蒋兆贯[1] XIA Peng;WANG Guoxiang;LIU Dong;JIANG Zhaoguan(Department of Radiology,Wuxi City Hospital of Traditional Chinese Medicine,Wuxi 214071,Jiangsu Province,China)

机构地区：[1]无锡市中医医院放射科

出　　处：《肿瘤影像学》2019年第5期319-324,共6页Oncoradiology

基　　金：江苏省无锡市科技局项目(CSZ00941)

摘　　要：目的:探讨常规磁共振成像(magnetic resonance imaging,MRI)图像纹理分析对肝血管瘤和原发性肝癌的鉴别诊断价值。方法:应用Mazda软件分析59例患者(共128个病灶)的磁共振T1WI和T2WI图像纹理。分别采用Fisher系数(Fisher coefficient,Fisher)、分类错误概率联合平均相关系数(classification error probability combined with average correlation coefficients,POE+ACC)、交互信息(mutual information,MI)和3种方法联合等纹理特征提取方法提取肝局灶性病变的纹理特征;运用原始数据分析(raw data analysis,RDA)、线性判别分析(linear discriminant analysis,LDA)、非线性判别分析(nonlinear discriminant analysis,NDA)和主要成分分析(principal component analysis,PCA)等4种统计学方法进行判别分类,结果以错判率表示。用非参数检验方法Mann-Whitney U检验比较两种成像序列图像纹理特征,判别分类的误判率差异。结果:基于T1WI图像纹理分析分类判别原发性肝癌和肝血管瘤的误判率在9.38%~35.16%,基于T2WI图像纹理分析分类判别的误判率在0.00%~26.56%,两组之间差异有统计学意义(P=0.002)。结论:常规MRI纹理分析可为肝血管瘤和原发性肝癌的鉴别诊断提供可靠的客观依据。Objective:To investigate the feasibility of texture analysis for the classification of hemangioma and primary liver cancer on conventional magnetic resonance(MR)T1-and T2-weighted images.Methods:Fifty-nine patients total of 128 lesions with hemangioma or primary liver cancer were enrolled in this study.Conventional MR T1-and T2-weighted images were available for further analysis.Texture features were calculated by using Mazda software.Fisher coefficient(Fisher),classification error probability combined with average correlation coefficients(POE+ACC),and mutual information(MI)were used to extract subsets of optimized texture features.Raw data analysis(RDA),linear discriminant analysis(LDA),nonlinear discriminant analysis(NDA)and principal component analysis(PCA)were used for lesion classification.The results are indicated with misclassification rate.The differences of the misclassification rate based on T1-and T2-weighted image texture were analyzed by Mann-Whitney U test.Results:Texturebased classification produced misclassification rates of 9.38%-35.16%on conventional magnetic resonance T1-weighted images and 0.00%-26.56%on T2-weighted images.There was statistically significant difference of the misclassification rate between the texture analysis based on T1-and T2-weighted images(P=0.002).Conclusion:Texture-based classification of hemangioma and primary liver cancer is feasible on conventional magnetic resonance images.

关 键 词：磁共振成像 肝局灶性病变 肝血管瘤 原发性肝癌 纹理分析 

分 类 号：R735.7[医药卫生—肿瘤] R445.2[医药卫生—临床医学]