目的本研究旨在评估瘤内和瘤周影像组学模型对肝细胞癌(hepatocellular carcinoma,HCC)微血管侵犯(microvascular invasion,MVI)的预测价值.材料与方法回顾性分析2016年至2023年间在三家医院经手术病理证实为HCC患者的钆塞酸二钠(Gd-EOB-DTPA)增强MRI图像,以及性别、年龄、肿瘤最大径、甲胎蛋白(alpha-fetoprotein,AFP)、丙氨酸转氨酶(alanine aminotransferase,ALT)、天冬氨酸转氨酶(aspartate aminotransferase,AST)和有无乙肝7项临床信息.在动脉期、门静脉期和肝胆期图像中勾画瘤内及瘤周5 mm、10 mm感兴趣区域(region of interest,ROI),从中提取影像组学特征;训练集中,运用多因素logistic回归分析筛选MVI的独立临床预测因素;应用支持向量机(support vector machine,SVM)建立瘤内模型、瘤周模型、瘤内联合瘤周模型、临床模型和临床影像组学模型共10种模型,采用受试者工作特征(receiver operating characteristic,ROC)曲线评估模型的预测能力,以DeLong检验比较各模型ROC曲线下面积(area under the curve,AUC)的差异.结果肿瘤最大直径[优势比(odds ratio,OR):1.449,95％置信区间(confidence interval,CI):1.212～1.733]及AFP(OR:1.645,95％CI:0.665～4.071)是基于训练集的MVI独立临床预测因子.验证集中,临床模型、瘤内模型、瘤周模型、瘤内+瘤周模型及临床影像组学模型预测HCC MVI的AUC值分别为0.728、0.764～0.820、0.791～0.795、0.804～0.828和0.747,瘤内+瘤周5 mm模型、瘤内+瘤周10 mm模型的AUC值分别为0.828(95％CI:0.728～0.929)、0.804(95％CI:0.696～0.913).各模型中,瘤内+瘤周5 mm模型与临床模型、临床影像组学模型的AUC差异具有统计学意义(P=0.039,0.028),其余模型间的AUC差异均无统计学意义(P>0.05).结论基于Gd-EOB-DTPA增强MRI影像组学模型可用于术前预测HCC MVI,其中瘤内+瘤周5 mm模型对HCC MVI有较高的预测能力,可为临床制订个体化治疗方案提供依据.
Clinical application value of predicting microvascular invasion in hepatocellular carcinoma using intratumoral and peritumoral radiomics models:A multicenter study
Objective:The aim of this study was to evaluate the predictive value of intratumoral and peritumoral radiomics models for microvascular invasion (MVI) in hepatocellular carcinoma (HCC). Materials and Methods:Gadoxetic acid disodium (Gd-EOB-DTPA) enhanced MRI images of patients with surgically pathologically confirmed HCC at three hospitals between 2016 and 2023 were retrospectively analyzed,as well as seven clinical information,including gender,age,maximum tumor diameter,alpha-fetoprotein (AFP),alanine aminotransferase (ALT),aspartate aminotransferase (AST),and the presence or absence of hepatitis B. Intratumoral regions and 5 mm and 10 mm peritumoral regions of interest (ROI) were outlined in arterial phase images,portal venous phase images,and hepatobiliary phase images,from which radiomics features were extracted;in the training cohort,multifactorial logistic regression analysis was applied to screen independent clinical predictors of MVI;support vector machine (SVM) was applied to establish a total of 10 models including intratumoral models,peritumoral models,intratumoral combined peritumoral models,clinical model,and clinical-radiomics combined model. Receiver operating characteristic (ROC) curve was used to evaluate the diagnostic efficacy of the models and DeLong test was employed to compare the difference of area under the curve (AUC). Results:Maximum tumor diameter[dominance ratio (OR):1.449,95％ confidence interval (CI):1.212-1.733]and AFP (OR:1.645,95％ CI:0.665-4.071) were independent clinical predictors of MVI based on the training cohort. In the validation cohort,the AUCs of the clinical model,intratumoral models,peritumoral models,intratumoral plus peritumoral models,and clinical-radiomics combined model for predicting MVI of HCC were 0.728,0.764-0.820,0.791-0.795,0.804-0.828,and 0.747,respectively,and those of the intratumoral plus 5 mm peritumoral model,intratumoral plus 10 mm peritumoral model were 0.828 (95％ CI:0.728-0.929),0.804 (95％ CI:0.696-0.913). Among the models,the AUC of the intratumoral plus 5 mm peritumoral model was statistically different from that of the clinical model and the clinical-radiomics combined model (P=0.039,0.028),and the differences in the AUCs among the rest of the models were not statistically significant (P>0.05). Conclusions:The Gd-EOB-DTPA-based enhanced MRI radiomics models can be used for preoperative prediction of HCC MVI,in which the intratumoral plus 5 mm peritumoral model has a higher predictive ability for HCC MVI. This model helps in the development of individualized treatment.
国家科技期刊平台
NSTL
万方数据
