基于高分辨率T_(2)WI影像组学联合临床特征预测食管癌新辅助放化疗疗效的价值

The value of predicting the efficacy of neoadjuvant radiotherapy and chemotherapy for esophageal cancer based on high-resolution T_(2)WI radiomics and clinical features

目的:探讨基于高分辨率(HR)T_(2)WI影像组学联合临床特征预测食管癌新辅助放化疗后疗效的价值。方法:回顾性分析本院2016年1月-2021年12月新辅助放化疗前接受HR T_(2)WI成像检查并经病理证实的95例食管癌患者资料。依据新辅助放化疗后病理缓解状态结果将患者疗效分为缓解组和未缓解组,在HR T_(2)WI图像上勾画肿瘤感兴趣区(ROI)后采用A.K软件提取影像组学特征,采用最大相关最小冗余(mRMR)算法进行降维,采用逻辑回归模型对筛选出的影像组学特征及临床参数构建模型;采用受试者操作特征(ROC)曲线评估不同模型的预测效能,计算曲线下面积(AUC)、准确率、敏感度和特异度,并采用DeLong检验比较不同模型预测食管癌新辅助放化疗敏感性的效能。结果:缓解组与未缓解组年龄差异具有统计学意义(P=0.001),其他临床特征差异无统计学意义(P>0.05)。从1688个组学特征中逐层筛选出4个影像组学特征,构建两个预测模型:基于HR T_(2)WI的影像组学模型、年龄-影像组学模型。HR T_(2)WI影像组学模型在训练集与验证集预测食管癌新辅助放化疗是否缓解的AUC、准确率、敏感度和特异度分别为0.863、80.0%、88.2%、76.7%,0.809、81.5%、75.0%、84.2%。年龄-影像组学模型在训练集与验证集预测食管癌新辅助放化疗是否缓解的AUC、准确率、敏感度和特异度分别为0.888、81.7%、94.1%、76.7%,0.836、81.5%、87.5%、78.9%。年龄-影像组学模型预测食管癌新辅助放化疗是否缓解的AUC、准确率和敏感度均高于HR T_(2)WI影像组学模型,而两者的特异度相仿。结论:基于HR T_(2)WI影像组学模型对食管癌新辅助放化疗是否缓解具有较好的预测效能,且HR T_(2)WI影像组学联合年龄特征模型显示出更高的预测价值。

Objective:To explore the value of predicting the efficacy of neoadjuvant radiotherapy and chemotherapy for esophageal cancer based on high-resolution T_(2)WI radiomics and clinical features.Methods:The data of 95 patients with esophageal cancer confirmed by pathology who underwent high-resolution T_(2)WI imaging before neoadjuvant radiotherapy and chemotherapy in our hospital from January 2016 to December 2021 was retrospectively analyzed.According to the results of pathological remission after neoadjuvant radiotherapy and chemotherapy,the patients were divided into remission group and no-remission group.After the tumor region of interest(ROI)was delineated on the high-resolution T_(2)WI image,the radiomics characteristics were extracted using A.K software,the maximum correlation minimum redundancy(mRMR)algorithm was used to reduce the dimension,and the logical regression model was used to build the model of the screened image group characteristics and clinical parameters.The predictive efficacy of different models was evaluated using the operating characteristic curve of subjects,and the area under the curve(AUC),accuracy,sensitivity and specificity were calculated.The efficacy of different models in predicting the sensitivity of neoadjuvant radiotherapy and chemotherapy for esophageal cancer was compared using DeLong test.Results:The clinical characteristics of the remission group and the no-remission group were tested for statistical differences,and only age was significant(P=0.001).The clinical characteristics of the other two groups were not significantly different(P>0.05).A total of 2 prediction models were obtained:based on high-resolution T_(2)WI radiomics model,and based on high-resolution T_(2)WI radiomics combined with age model.The AUC,accuracy,sensitivity and specificity of predicting whether to remission of neoadjuvant radiotherapy and chemotherapy for esophageal cancer based on the high-resolution T_(2)WI radiomics model in the training set and the validation set were 0.863,0.800,88.2%,76.7%,0.809,0.815,75.0%and 84.2%,respectively.The AUC,sensitivity and specificity of predicting whether to remission of neoadjuvant radiotherapy and chemotherapy for esophageal cancer in the training set and the validation set based on the high-resolution T_(2)WI radiomics combined with age feature model were 0.888,0.817,94.1%,76.7%,0.836,0.815,87.5%and 78.9%,respectively.The AUC,accuracy,and sensitivity of predicting whether to remission of neoadjuvant radiotherapy and chemotherapy for esophageal cancer based on high-resolution T_(2)WI radiomics combined with age characteristics model were higher than those of single high-resolution T_(2)WI radiomics model.And the specificity of the two was similar.Conclusion:Radiomics model based on the HR T_(2)WI has a good predictive effect on the response of esophageal cancer to neoadjuvant radiotherapy and chemotherapy,and the combination of high-resolution T_(2)WI radiomics and age characteristic model shows a higher predictive value.