摘要
目的探讨基于平扫与增强CT图像的纹理特征在评价肝硬化肝脏质地模型中的价值。方法回顾性选取100例临床确诊肝硬化患者的肝脏平扫加增强CT图像作为研究组;收集98例同时期无弥漫性肝病患者的肝脏平扫加增强CT图像作为对照组。图像分割由两名医生应用Shukun Radiomics V94软件对选定层面图像进行ROI勾画,提取纹理特征,通过降维处理,保留高度可重复的9个最显著特征,经Logistic回归分析,分别建立基于平扫及增强CT图像纹理特征的肝硬化质地模型,并验证各模型的诊断效能。应用SPSS的卡方检验及SAS软件宏命令包对平扫与增强CT图像纹理特征的曲线下面积(AUC)值、特异度、敏感度、符合率及诊断效能进行统计学分析。结果基于平扫及增强CT图像纹理特征建立的肝硬化患者肝脏质地模型均能准确诊断肝硬化,模型AUC值均在0.95以上。平扫图像模型训练集AUC值(0.9617)、符合率(0.9312)、特异度(0.8769)略低于增强图像模型训练集AUC值(0.9919)、符合率(0.9771)、特异度(0.9846);而敏感度(0.9848)略高于增强图像模型(0.9697)。平扫图像模型验证集AUC值(0.9501)、符合率(0.9254)、特异度(0.9412)及敏感度(0.9275)均略低于增强图像模型验证集AUC值(0.9935)、符合率(0.9701)、特异度(1.0000)及敏感度(0.9412)。增强CT纹理特征建立的肝硬化肝脏质地模型AUC值略高于平扫CT纹理特征建立的模型,但两者间无统计学差异,P>0.05(0.0875)。结论肝脏平扫或增强CT图像纹理特征均可有效评估肝硬化患者肝脏质地改变,增强CT模型的AUC略高于平扫CT模型。
Objective To investigate the value of texture features based on unenhanced and enhanced CT images in the liver texture model of cirrhotic patients.Methods The unenhanced and enhanced CT images of 100 patients with clinically confirmed liver cirrhosis and 98 patients without diffuse liver disease were retrospectively analyzed.Shukun Radiomics V94 software was used to draw regions of interest(ROI)on selected layer images by two radiologists,and texture features were extracted.After dimension reduction,9 most significant features were retained and logistic regression analysis was performed to establish texture models of cirrhosis.The diagnostic efficiency of each model was verified.Area under receiver operating characteristic curve(AUC)value,diagnostic accuracy,specificity,and sensitivity of the texture features were compared.Results The cirrhotic liver texture models extracted from both unenhanced and enhanced CT image features could accurately diagnose cirrhosis with AUC values above 0.95.The AUC value(0.9617),diagnostic accuracy(93.12%),and specificity(87.69%)of the unenhanced image model training set were slightly lower and the sensitivity(98.48%)was slightly higher than those of the enhanced image model training set(0.9919,97.71%,98.46%,96.97%).The AUC value(0.9501),accuracy(92.54%),specificity(94.12%),and sensitivity(92.75%)of the unenhanced image model verification set were slightly lower than those of the enhanced image model verification set(0.9935,97.01%,100%,94.12%).The AUC value of the liver texture model of cirrhosis established by enhanced CT texture features was not significantly higher than that established by unenhanced CT texture features(P>0.05).Conclusion Both unenhanced and enhanced CT texture models can evaluate the texture changes of liver in cirrhotic patients effectively.The optimal selection should be made according to the clinical situation.
作者
李民
张喆
赵丽琴
杨大为
刘长春
荆利娜
闫玉昌
常泰
LI Min;ZHANG Zhe;ZHAO Li-qin;YANG Da-wei;LIU Chang-chun;JING Li-na;YAN Yu-chang;CHANG Tai(Department of Radiology,Beijing Tiantan Hospital,Capital Medical University,Beijing 100070,China)
出处
《影像诊断与介入放射学》
2021年第5期347-352,共6页
Diagnostic Imaging & Interventional Radiology
基金
北京市自然科学基金项目(7192042)。
