CT影像组学预测巴塞罗那临床肝癌分期系统早中期肝细胞癌的可行性研究

A feasibility study of CT radiomics for predicting hepatocellular carcinoma at early and intermediate stages

目的:探讨CT影像组学预测巴塞罗那临床肝癌分期系统(BCLC)早中期肝细胞癌的可行性。方法:回顾性分析190例BCLC分期中早中期肝细胞癌患者的CT平扫、增强扫描动脉期及门静脉期图像。采用ITK-SNAP软件手动勾画病灶ROI;用Anaconda软件提取ROI纹理特征,采用最小绝对收缩与选择算子(LASSO)回归模型降维运算获取纹理特征,并建立预测模型。通过ROC曲线评价预测模型的检测效能。结果:基于CT平扫图像的影像组学模型在测试集诊断BCLC 0期(最早期)、A期(早期)及B期(中期)的AUC分别为0.61、0.51、0.55,总准确率为35.4%;基于动脉期图像影像组学模型在测试集诊断0、A及B期的AUC分别为0.99、0.98、0.99,总准确率为92.7%;基于门静脉图像的影像组学模型在测试集诊断0、A及B期的AUC分别为0.98、0.95、0.99,总准确率为90.9%。DeLong检验显示,动脉期组学模型及门静脉期组学模型均与平扫组学模型对A、B期肝细胞癌诊断效能差异均有统计学意义(均P<0.05),动脉期组学模型与门静脉期组学模型对0、A、B期的诊断效能差异均无统计学意义(均P>0.05)。结论:动脉期及门静脉期的CT影像组学模型可用于预测BCLC早中期肝细胞癌。

Objective:To explore the feasibility of CT radiomics for predicting hepatocellular carcinomas at early and intermediate stages of Barcelona Clinic Liver Cancer(BCLC)system.Methods:CT images of 190 patients with BCLC early and intermediate stages hepatocellular carcinoma were retrospectively analyzed.ITK-SNAP software and Anaconda software were used to construct ROI and extract the texture features.Least absolute shrinkage and selection operator(LASSO)regression analysis was used for feature selection and establishment of prediction model.ROC curve was used to analyze the diagnostic effectiveness.Results:The AUCs of the prediction model of the non-enhanced CT images for BCLC earliest,early and intermediate stages in the validation cohort were 0.61,0.51 and 0.55,those of the arterial phase images were 0.99,0.98 and 0.99,those of the portal vein phase images were 0.98,0.95 and 0.99,and the total accuracy rate of the three models was 35.4%,92.7%and 90.9%,respectively.DeLong test showed that the effectiveness of the prediction models of the arterial and portal vein phase images was different with that of the non-enhanced CT images in the diagnosis of early and intermediate stages hepatocellular carcinoma(both P<0.05),while there was no difference between the arterial phase images and the portal vein phase images(P>0.05).Conclusion:Contrast-enhanced CT radiomics model can be used to predict hepatocellular carcinoma at BCLC early and intermediate stages.