术前CT影像组学预测肝细胞癌微血管侵犯  

Prediction of Microvascular Invasion in Hepatocellular Carcinoma by Preoperative CT Radiomics

作　　者：韦武鹏[1] 覃云英 廖海 廖玉婷 谭丽芬 卢金丹 苏丹柯[2,3,4] WEI Wupeng;QIN Yunying;LIAO Hai(Department of Radiology,The Second Affiliated Hospital of Guangxi Medical University,Nanning,Guangxi Province 530007,P.R.China)

机构地区：[1]广西医科大学第二附属医院放射科,南宁530007 [2]广西医科大学附属肿瘤医院影像中心 [3]广西临床重点专科(医学影像科) [4]广西医科大学附属肿瘤医院优势培育学科(医学影像科),530021 [5]通用电气药业(上海)有限公司,广州510623

出　　处：《临床放射学杂志》2021年第12期2378-2385,共8页Journal of Clinical Radiology

基　　金：广西重点研发计划项目(编号:桂科AB18126031);广西影像医学临床医学研究中心建设项目(编号:桂科AD20238096)。

摘　　要：目的探讨术前CT影像组学在肝细胞癌微血管侵犯(MVI)预测中的应用价值。方法回顾性分析219例孤立性肝细胞癌患者术前临床及术前CT资料。患者术前增强CT的门静脉期图像提取1361个影像组学特征。构建基于临床+影像图像分析特征的Clinical模型、影像组学特征的Feature模型和合并上述特征的Comb模型。通过Hosmer-Lemeshow检验对模型进行验证,ROC曲线、AUC、敏感性、特异性和DeLong检验对模型进行评价,并构建列线图。结果 Hosmer-Lemeshow检验结果显示3个模型模型拟合优良,在训练队列和测试队列均实现了较好的预测效能。包含影像组学特征的Comb和Feature模型在训练队列均显著优于Clinical模型,在测试队列无显著差异。决策曲线分析显示Comb模型和Feature模型在大部分阈值概率的临床获益高于Clinical模型。结论基于CT门静脉期的影像组学特征具有术前预测肝细胞癌MVI的潜力。Objective To evaluate the value of preoperative CT imaging in predicting microvascular invasion of hepatocellular carcinoma. Methods The clinical and preoperative CT data of 219 patients with isolated HCC were analyzed retrospectively. 1361 radiomics features were extracted from preoperative CT portal vein images of all patients. Construct a Clinical model based on clinical + image analysis features,a Feature model based on radiomics features,and a Comb model combining the above features. Hosmer-Lemeshow test,ROC curve,AUC,sensitivity,specificity and De Long test were used to verify and evaluate these models,and nomograms were constructed. Results Hosmer-Lemeshow test showed that these models fit well and achieved good prediction performance in the training group and the test group. The Comb and Feature models including radiomics characteristics were significantly better than the Clinical model in the training group,but there was no significant difference in the test group. According to decision curve analysis,the linical benefit of Comb model and Feature model were higher than that of Clinical model in most probability. Conclusion The radiomics features based on CT portal venous phase images of hepatocellular carcinoma have the potential to predict MVI before operation.

关 键 词：肝细胞癌 微血管侵犯 影像组学 体层摄影术 X线计算机 

分 类 号：R735.7[医药卫生—肿瘤] R730.44[医药卫生—临床医学]