基于增强门静脉期CT提取影像组学特征构建Nomogram评估肝纤维化分级  

Developmenta Nomogram Based on Radiomics Extracted from Enhanced-CT to Predict Staging of Liver Fibrosis

作　　者：张斯佳[1] 祁佩红[1] 杨新焕[1] 张伟[1] 石俊英[1] ZHANG Sijia;QI Peihong;YANG Xinhuan(Medical Imaging Department,Zhengzhou People's Hospital,Zhengzhou,Henan Province 450003,P.R.China)

机构地区：[1]郑州人民医院医学影像科,450003

出　　处：《临床放射学杂志》2021年第12期2319-2325,共7页Journal of Clinical Radiology

基　　金：2018年河南省医学科技计划(联合共建)项目(编号:2018020831)。

摘　　要：目的探究基于增强CT提取影像组学特征构建Nomogram无侵袭性评估肝纤维化分级的可行性。方法纳入郑州市人民医院2015年2月~2019年2月入院确诊为肝纤维化的患者,分为肝纤维化和早期肝硬化。以2018年2月为时间截点,将该时间点之前的入组患者分为训练集,将该时间点之后的入组患者分为测试集,并基于患者增强CT图像提取影像组学特征。以患者是否肝纤维化或早期肝硬化为研究标签,对训练集患者采用最小冗余最大相关(mRMR)进行影像组学特征去冗除杂,继而采用套索算法(LASSO)构建影像组学标签(Radscore)。联合临床特征与Radscore构建临床影像联合模型评估肝纤维化分级。结果共计纳入91例,其中肝纤维化患者59例,早期肝硬化患者35例,基于门静脉期CT图像共计提取845个纹理特征,经过特征降维利用16个影像组学特征构建影像组学标签Radscore, Radscore鉴别训练集和测试集的患者是否肝纤维化或肝硬化早期的曲线下面积(AUC)为0.885 vs 0.932。继而基于训练集患者的临床特征联合Radscore构建多元逻辑回归模型评估患者是否肝纤维化或肝硬化早期,Nomogram可视化影像临床联合模型,训练集中Radscore的AUC高于临床影像模型高于临床指数[0.885 (95%CI:0.803~0.965) vs 0.881 (0.800~0.961) vs 0.672 (0.532~0.810)],测试集中临床影像模型的AUC高于Radscore高于临床指数[0.0.924 (0.812~0.1.000) vs 0.932 (0.821~0.993) vs 0.552 (0.315~0.783)]。Hosmer-Lemeshow分析提示联合模型与实际观察情况差异无统计学意义(P=0.8514)。结论基于门静脉期CT提取影像组学参数联合临床信息可用于无侵袭性评估肝纤维化分级,协助临床上针对不同患者制定个体化治疗方案。Objective To explore the predictionofa nomogram constructed by radiomics based enhanced CT for liver fibrosis staging. Methods Patients admitted to Zhengzhou People’s Hospital from February 2015 to February 2019 diagnosed with liver fibrosis were included and divided into liver fibrosis and early cirrhosis.Taking February 2018 as the time cut-off point, patients enrolled before this time point were divided into a training set, and patients enrolled after this time point were divided into a test set.Radiomics features were extracted based on patients’ enhanced CT images.Taking whether patients had liver fibrosis or early cirrhosis as the research label, we used max-relevance and min-redundancy(mRMR) to remove the Redundancy and Redundancy of the imaging features of the patients in the training set.Subsequently, Radiomics signature(Radscore) is constructed by least absolute selection operator(LASSO).Combined clinical features and Radscore were used to construct a combined clinical image model to evaluate the grade of liver fibrosis. Results A total of 91 people were included in this retrospective study, including 59 patients with liver fibrosis and 35 patients with early cirrhosis.A total of 845 texture features were extracted based on portal phase CT images, and 16 Radiomics features were used to construct the Radiomics label Radscore after feature demolification.The Radscore AUC for differentiating between patients in the training set and test set from those in the early stage of liver fibrosis or cirrhosis was 0.885 vs 0.932.Then, a multiple logistic regression model was constructed based on the clinical characteristics of the patients in the training set combined with Radscore to evaluate whether the patients were in the early stage of liver fibrosis or cirrhosis.For the Nomogram visual imaging combined clinical model, the AUC of the Radscore in the training set was higher than that of the clinical imaging model and the clinical index [0.885(95%CI:0.803-0.965) vs 0.881(0.800-0.961) vs 0.672(0.532-0.810)],Th

关 键 词：肝纤维化 肝硬化 增强CT 影像组学 分级 

分 类 号：R575.2[医药卫生—消化系统] R816.5[医药卫生—内科学]