基于影像组学的食管鳞状细胞癌放疗后生存预测模型的建立与验证

  目的  利用增强CT图像特征构建模型，预测食管鳞状细胞癌患者放疗后原发病灶的局部控制情况。  方法  2016年7月至2017年12月218例于河北医科大学第四医院接受放化疗且病理诊断为鳞状细胞癌的食管癌患者随机分为训练组（153例）和验证组（65例），提取训练组患者的增强CT图像影像组学特征，构建并验证模型预测食管癌患者放疗后原发病灶局部控制的效能。采用ROC曲线、C-index曲线、校准曲线和决策曲线评价不同模型的性能。  结果  在训练组筛选出6个有意义的影像组学特征，构建预测食管鳞状细胞癌放疗后原发病灶局部控制的放射学标签。训练组和验证组的ROC曲线下面积分别为0.758和0.728；C-index为0.709和0.695；以放射学标签得分?0.22为界值分为高危组和低危组，低危组患者的1、3、5年无原发病灶复发生存率均高于高危组患者（P<0.05）。结合临床因素与放射学标签构建食管癌放疗后无原发病灶复发生存的影像组学列线图模型，ROC曲线评价预测效能的曲线下面积在训练组和对照组分别为0.775和0.740；C-index分别为0.722和0.707；以影像组学列线图模型得分0.55为界值分为高危组和低危组，低危组患者的1、3、5年无原发病灶复发生存率均高于高危组患者（P<0.05）。  结论  成功构建预测食管癌放疗后无原发病灶复发生存的模型，模型具有较好的临床预测价值。 

Construction and validation of a radiomic-based survival prediction model for patients with esophageal squamous cell carcinoma after radiotherapy

  Objective  To assess the clinical predictive value of a model based on enhanced CT image features to enable the local control of post-radiotherapy primary lesions in patients with esophageal squamous cell carcinoma.  Methods  Data from 218 patients with esophageal squamous cell carcinoma, diagnosed by pathology and receiving radical radiotherapy and chemotherapy, were collected between July 2016 and December 2017 at The Fourth Hospital of Hebei Medical University. Patients were randomly assigned into training (n=153) and validation groups (n=65). Imaging features were extracted after delineation of the region of interest (ROI) from enhanced CT images captured before and after radiotherapy. The radiomics features of the training group in combination with the clinical features were used to construct a radiomics nomogram to predict the local control of primary lesions after radiotherapy. Receiver operating characteristic (ROC) curves, C-index curves, calibration curves, and decision curves were used to evaluate the performances of the different models.  Results  Six significant radiomic features were extracted from the training group to generate radiological labels for the prediction of local control of post-radiotherapy primary lesions in esophageal cancer. In the training and validation groups, the areas under the ROC curves were 0.758 and 0.728, respectively, and their C-indices were 0.709 and 0.695, respectively. Patients were classified as high- and low-risk based on a radiological label score, whose cut-off was set at -0.22. The 1-, 3-, and 5-year survival rates of the patients classified a slow-risk were higher than those of the patients classified as high-risk (P < 0.05). Additionally, a radiomics nomogram was generated to predict the survival of patients with esophageal cancer without any recurrence of the primary lesion after radiotherapy. The area under the curve of the nomogram was 0.775 in the training group and 0.740 in the validation group, while the corresponding C-indices were 0.722 and 0.707, respectively. A radiomics nomogram model score of 0.55 was used as a cut-off to classify patients into high-and low-risk groups. Patients in the low-risk group had higher 1-, 3-, and 5-year survival rates coupled with the absence of primary lesion recurrence compared to those in the high-risk group (P< 0.05).  Conclusions  A radiological labeling method and radiomics nomogram were successfully constructed to predict survival without primary lesion recurrence among patients with esophageal cancer after radiotherapy. Both tools were found to have good clinical predictive value.