基于增强MRI的影像组学术前预测肝细胞癌微血管侵犯的价值  

The value of contrast enhanced MRI-based radiomics for preoperatively predicting microvascular invasion of hepatocellular carcinoma

作　　者：林涛 刘爱连[1,4] 赵莹 郭妍[2] 华正宇 祁文静 任雪[1,4] 许岂豪[1,4] 宋清伟 王楠 李昕[2] 吴艇帆 LIN Tao;LIU Ailian;ZHAO Ying;GUO Yan;HUA Zhengyu;QI Wenjing;REN Xue;XU Qihao;SONG Qingwei;WANG Nan;LI Xin;WU Tingfan(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Dalian,Liaoning Province 116011,China;GE Healthcare,Shanghai 200000,China;Department of Pathology,the First Affiliated Hospital of Dalian Medical University,Dalian,Liaoning Province 116011,China;Dalian Engineering Research Center for Artificial Intelligence in Medical Imaging,Dalian,Liaoning Province 116011,China)

机构地区：[1]大连医科大学附属第一医院放射科,辽宁大连116011 [2]通用电气医疗集团,上海200000 [3]大连医科大学附属第一医院病理科,辽宁大连116011 [4]大连市医学影像人工智能工程技术研究中心,辽宁大连116011

出　　处：《实用放射学杂志》2022年第4期562-567,共6页Journal of Practical Radiology

基　　金：国家自然科学基金面上项目(61971091).

摘　　要：目的探讨基于增强MRI的影像组学术前预测肝细胞癌(HCC)微血管侵犯(MVI)的价值.方法回顾性选取行1.5T MR检查且经病理证实的116例HCC患者,其中MVI(+)51例,MVI(-)65例,按7︰3比例随机分为训练集80例[MVI(+)35例,MVI(-)45例]和测试集36例[MVI(+)16例,MVI(-)20例].由2名放射科医师分别在动脉期、门静脉期及延迟期MRI图像上逐层手动勾画病灶轮廓并生成三维感兴趣区(ROI),各个期相分别提取107个影像组学特征.利用组内相关系数(ICC)、Spearman相关性检验及梯度提升决策树(GBDT)筛选特征,利用筛选后的特征建立相应影像组学模型,并从中选择最优模型进行后续分析.采用单因素及多因素逻辑回归方法筛选最有意义的临床及常规影像学特征并建立传统模型.采用逻辑回归构建结合了影像组学特征、临床及常规影像学特征的联合模型.采用受试者工作特征(ROC)曲线和决策曲线分析(DCA)评价各模型的效能和临床应用价值.结果传统模型、增强3期联合组学模型及联合模型在测试集中的曲线下面积(AUC)分别为0.614、0.828及0.841;联合模型的AUC较传统模型有显著意义的提升(AUC:0.841和0.614,P=0.0099),在测试集中的敏感度及特异度分别为68.8%及80%.结论基于增强MRI的影像组学有助于在术前预测HCC的MVI;联合模型是最优的MVI预测模型,能以无创的方式为HCC临床诊疗方案的制订提供重要的参考信息.Objective To explore the value of contrast enhanced MRI-based radiomics for preoperatively predicting microvascular invasion(MVI)of hepatocellular carcinoma(HCC).Methods 116 patients with HCC confirmed by pathology were retrospectively collected,including 51 cases with MVI(+)and other 65 cases with MVI(-).All patients underwent 1.5T MR scan.Those cases were randomly divided into the training sets 80 cases[including 35 cases for MVI(+)and 45 cases for MVI(-)]and test sets 36 cases[including 16 cases for MVI(+)and 20 cases for MVI(-)]at a 7:3 ratio.The contours of lesions were manually drawn on the MRI of arterial,portal vein and delayed phases by two radiologists and the three dimensional region of interest(ROI)was generated.107 radiomics features were extracted from each phase.Interclass corre elation coefficient(1CC),Spearman correlation test t;and gradient boosting decision tree(GBDT)were used to select the features and to construct radiomics models,and the optimal model was selected for subsequent analysis.Univariate and multivariate Logistic regression methods were used to screen significant clinical and imaging features,and then to establish the traditional model.Logistic regression was used to construct a combined model incorporating the radiomics model and traditional model.Receiver operating characteristic(ROC)curve and decision curve analysis(DCA)were used to analyze the efficacy and clinical application value of three models.Results The area under the curve(AUC)of the traditional model,the enhanced-triphase-combined radiomics model,and whole combined model were 0.614,0.828 and 0.841 in the test sets,respectively.The predictive efficiency of the proposed combined models was significantly improved when compared with the traditional model(AUC:0.841 and 0.614,P=0.0099),and the sensitivity and specificity in the test set were 68.8%and 80%,respectively.Conclusion The radiomics models based on contrast enhanced MRI can help to predict the MVI status of HCC and the combined models show better predictive capabil

关 键 词：肝细胞癌 微血管侵犯 影像组学 磁共振成像 

分 类 号：R735.7[医药卫生—肿瘤] R445[医药卫生—临床医学] R445.2