基于术前MRI影像组学特征预测肝癌经肝动脉化疗栓塞治疗的反应性

目的 探索术前MRI影像组学特征预测肝细胞癌（hepatocellular carcinoma，HCC）对经肝动脉化疗栓塞（transcatheter arterial chemoembolization，TACE）治疗反应性的可行性。方法 回顾性收集2016年3月至2018年4月经临床或病理确诊、并在温州医科大学附属第五医院接受TACE治疗的中晚期HCC患者86例，采用随机数字表法分为训练组（n=59）及验证组（n=27），基于ITK-SNAP软件在术前T2WI图像勾画肿瘤感兴趣区，使用GE AI-Kit软件提取影像特征；经最小绝对收缩与选择算子（LASSO）回归筛选最优的特征子集，并构建TACE反应性预测模型；利用受试者工作特征（ROC）曲线、校准曲线及决策曲线分析评价模型的诊断效能和临床应用价值。结果 共筛选出8个影像学特征与TACE反应性相关，并成功构建预测模型。训练组中，该模型的ROC曲线下面积（AUC）为0.838（95%CI 0.737～0.939），灵敏度为74.1%，特异度为84.4%；在验证组中，AUC为0.794（95%CI 0.604～0.985），灵敏度为75.0%，特异度为80.0%。校准曲线显示该模型在训练组和验证组的预测概率与实际概率拟合较好；决策曲线分析显示该模型在0.1～1.0的阈值范围内具有较高的净收益。结论 术前MRI T2WI图像影像组学特征可用于预测HCC的TACE 反应性，基于MRI T2WI图像影像特征的预测模型特异度和灵敏度较高。

Predict the responsiveness of transcatheter arterial chemoembolization for hepatocellular carcinoma based on preoperative MRI imaging features

Objective To explore the feasibility of preoperative MRI imaging features to predict the response of hepatocellular carcinoma (HCC) to transcatheter arterial chemoembolization (TACE). Methods A total of 86 HCC patients who underwent TACE between Mar. 2016 to Apr. 2018 in the Fifth Affiliated Hospital of Wenzhou Medical University. Patients were divided into training cohort (n=59) and validation cohort (n=27) by a random number table method. ITK-SNAP software was used to delineate the tumor region of interest (ROI) on the preoperative MRI T2WI images. GE AI-Kit software was applid to extract radiomics features. The least absolute shrinkage and selection operator (LASSO) regression was used to select the optimal feature subset, which were integrated into a radiomics-based prediction model. The receiver operating characteristic (ROC) curve, nomogram, calibration curve and decision curve analysis (DCA) were used to evaluate the diagnostic efficacy and clinical useful of the model. Results A total of 8 radiomics features related to TACE response were selected and further integrated into the TACE response prediction model. In the training cohort group, the area under the ROC curve (AUC) was 0.838 (95%CI 0.737~0.939), the sensitivity was 74.1%, and the specificity was 84.4%. In the validation cohort group, the AUC was 0.794 (95%CI 0.604~0.985), the sensitivity was 75.0%, and the specificity was 80.0%. The calibration curve showed that the predicted probability of the model fits well with the actual probability. DCA showed that the model had a higher net benefit within the threshold range of 0.1~1.0. Conclusion Radiomics features based preoperative MRI T2WI images can be used to predict tumor response of HCC to TACE treatment. The prediction model based on radiomics features from preoperative MRI T2WI images has high specificity and sensitivity.