摘要
目的:探讨钆塞酸二钠(Gd-EOB-DTPA)增强MRI肝胆期影像组学对乙型肝炎病毒感染患者肝储备功能的评估价值.方法:将 73 例慢性乙型肝炎肝功能正常者、136 例Child-Pugh A级和 70例Child-Pugh B/C级乙型肝炎患者按照 8:2 随机分成训练组和测试组.从肝胆期MRI图像中勾画全肝实质进行影像组学特征提取.将得到的影像组学得分值(Rad-score)和多因素回归分析筛选出的临床变量联合,建立临床-影像组学联合模型.采用受试者工作特征(ROC)曲线分析评估效能.另一时间段 64 例(17 例肝功能正常、30 例 Child-Pugh A 级和 17 例 Child-Pugh B/C 级)乙肝患者用来验证.结果:多因素回归分析筛选出血小板、总胆红素和凝血酶原时间国际化比值用以建立鉴别肝功能正常与Child-Pugh A级的临床模型,利用 4 个最相关的影像组学特征构建影像组学模型,训练组、测试组和验证组中临床模型鉴别肝功能正常与 Child-Pugh A 级的曲线下面积(AUC)分别为 0.897、0.884、0.780,组学模型的AUC分别为 0.890、0.914、0.824,临床-影像组学联合模型的 AUC 分别为 0.951、0.978、0.886.联合模型与临床模型、组学模型相比,净重新分类改善指数(NRI)和综合判别改善指数(IDI)的检验值均＜0.05.血小板和总胆红素被筛选出构建鉴别 Child-Pugh A级与 Child-Pugh B/C 级的临床模型,7 个最相关的影像组学特征构建影像组学模型,训练组、测试组和验证组中影像组学模型鉴别Child-Pugh A级与 Child-Pugh B/C 级的 AUC 分别为 0.862、0.865、0.818,联合模型(AUC＝0.940、0.934、0.951)与临床模型(AUC＝0.916、0.893、0.914)的评估效能差异无统计学意义(P＞0.05).结论:基于Gd-EOB-DTPA增强MRI肝胆期影像组学对乙肝患者的肝储备功能具有良好的评估价值,联合临床变量后可以提高Child-Pugh A级的早期检出效能,但对于 Child-Pugh A 级与 Child-Pugh B/C 级的评估,单独的临床模型已经表现出了良好的效能.
Abstract
Objective:To investigate the value of radiomics based on the hepatobiliary phase ima-ges of Gd-EOB-DTPA enhanced liver MR in evaluating liver reserve function of hepatitis B virus(HBV)patients.Methods:73 chronic HBV patients with normal liver function,136 HBV patients with Child-Pugh A liver function and 70 patients with Child-Pugh B/C liver function were randomly divided into the training group and the test group in the ratio of 8:2.Radiomics features were extracted from the hepatobiliary phase MR images via delineating the whole liver parenchyma.The clinic-radiomics model combining the radiomics score(Rad-score)and clinical variables selected by multivariate re-gression analysis was built.Evaluation performances of the model were analyzed using receiver operat-ing characteristic(ROC)curve.64 HBV patients were used for validation from another period,inclu-ding 17 cases with normal liver function,30 with Child-Pugh A and 17 with Child-Pugh B/C.Results:Platelet(PLT),total bilirubin(TBIL)and international normalized ratio of prothrombin time(INR)were selected by multivariate regression analysis to construct clinical model for distinguishing normal liver function from Child-Pugh A.The radiomics model was constructed using the four most relevant radiomics features.The areas under the curve(AUCs)of clinical model for distinguishing normal liver function from Child-Pugh A were 0.897,0.884 and 0.780 in the training,test and validation cohorts,re-spectively.The AUCs of radiomics model were 0.890,0.914 and 0.824 in the training,test and valida-tion cohorts,respectively.The AUCs of clinic-radiomics combined model were 0.951,0.978 and 0.886 in the training,test and validation cohorts,respectively,whose net reclassification index(NRI)and in-tegrated discrimination improvement(IDI)were significantly different from the other two models(P＜0.05).PLT and TBIL were selected to construct clinical model for distinguishing Child-Pugh A from Child-Pugh B/C.The radiomics model was constructed using the seven most relevant radiomics features.The AUCs of radiomics model for distinguishing Child-Pugh A from Child-Pugh B/C were 0.862,0.865 and 0.818 in the training,test and validation cohorts,respectively.There was no significant difference between the clinic-radiomics combined model(AUC＝0.940,0.934 and 0.951,respectively)and the clinical model(AUC＝0.916,0.893 and 0.914,respectively)(P＞0.05).Conclusions:Radiomics based on the hepatobiliary phase images of Gd-EOB-DTPA enhanced MR has a high evaluation value for liver reserve function of HBV patients,which can improve the early detection performance of Child-Pugh A when combined with clinical variables.However,the single clinical model has shown good evaluation efficacy for Child-Pugh A vs.Child-Pugh B/C.
基金项目
江苏省"333高层次人才培养工程"项目(2022-3-6-139)
NETL
NSTL
维普
万方数据