基于DCE-CT的MaZda纹理分析技术在鉴别肝脏良恶性病变中的价值及最佳效能分析  

A study of optimal efficacy of MaZda texture analysis based on DCE-CT in the differential diagnosis between benign and malignant liver lesions

作　　者：韩瑞[1] 黄璐[2] 董进[1] 宋少辉[1] 蒋鸿[1] 尹龙[1] 彭勇[1] 张东友[1] HAN Rui;HUANG Lu;DONG Jin(Department of Radiology,the First Hospital of Wuhan,Wuhan 430022,China)

机构地区：[1]武汉市第一医院放射科,武汉430022 [2]华中科技大学同济医学院附属同济医院放射科,武汉430030

出　　处：《放射学实践》2019年第8期847-851,共5页Radiologic Practice

基　　金：武汉市卫计委面上重点项目(WX18B08)

摘　　要：目的:探讨基于CT动态增强的MaZda纹理分析技术在鉴别肝脏良恶性病变中的价值及不同分析方法组合的最佳效能。方法:回顾性分析经病理证实或临床动态随访确诊的101例肝脏良恶性病变(共112个病灶)患者的病例资料,其中良性组55例、恶性组57例。所有患者行CT三期动态增强扫描,使用纹理分析软件(MaZda)中的直方图、绝对梯度、游程矩阵、共生矩阵和自回归模型共5种分析方法,对每期图像上的肝脏病变进行纹理特征的提取,共获得256个纹理特征参数;采用Fiher相关系数、最小分类误判率+平均相关系数(POE+ACC)及互信系数(MI)三种统计学方法,分别筛选出鉴别肝脏良恶性病变的10个最佳纹理特征参数。使用B11程序中的主成分分析法(PCA)、线性鉴别分析法(LDA)和非线性鉴别分析法(NDA)对这10个最佳纹理参数进行降维和分类,计算不同期相、不同统计方法组合下这些最佳纹理特征参数鉴别肝脏良恶性病变的最小误判率(R)。结果:使用MaZda软件的纹理分析技术鉴别肝脏良恶性病变,基于CT增强扫描动脉期的最低误判率为15.18%,静脉期为13.39%,延迟期为11.61%,均在良好范围内(R≤20%)。鉴别能力与CT肝脏动态增强期相、纹理特征的统计学提取方法和降维方法的选择有关。无论动脉期、静脉期或延迟期,相对于其它组合方法,MI+NDA组合的鉴别诊断误判率最低;鉴别肝脏良恶性病变的最佳期相为延迟期,使用MI+NDA的组合鉴别诊断的误判率最低(13/112),主要误判的病灶包括肝脓肿4个、血管瘤4个、肝细胞癌3个和肝转移瘤2个。筛选提取的10个最佳纹理特征参数分别为90%百分位灰度值(Perc.90%)、均值(Mean)、Perc.50%、Perc.99%、Perc.10%、Perc.01%、差方差S(1,0)、熵S(4,-4)、熵S(2,0)和逆差矩S(2,2)。结论:基于CT动态增强的纹理分析技术在肝脏良恶性病变的鉴别诊断中是可行的;CT肝脏动态增强不同期相、纹Objective:To investigate the value of MaZda texture analysis technology based on dynamic contrast enhanced CT(DCE-CT)in differential diagnosis of benign and malignant liver lesions and the best efficiency of different combinations of analysis methods.Methods:101 patients with 112 benign and malignant liver lesions confirmed by pathology or clinical dynamic follow-up were studied retrospectively.All patients were divided into 2 groups,benign group(n=54)and malignant group(n=47).DCE-CT scan was performed on all patients.Texture analysis software(MaZda)was used to extract the texture features of the liver lesions in the two groups,with analysis model of histogram,absolute gradient,run-length matrix,co-occurrence matrix and autoregressive(AR).Totally 256 texture feature parameters was obtained.Then three statistic methods including Fisher coefficient,classification error probability with average correlation coefficients(POE+ACC),and mutual information coefficient(MI)in MaZda software were used to screen out 10 optimal texture features in each method group for differential diagnosis between the benign and malignant liver lesions.The principal component analysis(PCA),linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)were used to reduce dimensionality and classified for these texture parameters using B11 software in MaZda software.The minimum misdiagnosis rates(MMRs)of texture paremeters from different phases and different statistical methods in differential diagnosis were calculated.Results:Texture analysis technique using MaZda software was effective in differential diagnosis between the benign and malignant liver lesions,and the MMR of arterial phase,venous phase and delayed phase was 15.18%,13.39%and 11.61%,respectively.Its discriminant ability was related to the selection of dynamic enhanced phase,texture feature statistical extraction methods and texture feature dimensional reduction methods.Arterial phase,venous phase,delayed phase,and MI+NDA had lower MCR than those of other methods.The be

关 键 词：肝脏病变 体层摄影术 X线计算机 动态增强扫描 纹理分析 MaZda软件 

分 类 号：R814.42[医药卫生—影像医学与核医学] R735.7[医药卫生—放射医学]