CT影像组学标签预测局部进展期胃癌新辅助化疗疗效的多中心分析

A multicenter study of computed tomography-based radiomics signature prediction of neoadjuvant chemotherapy outcomes for locally advanced gastric cancer

目的通过多中心研究探讨基于CT图像的影像组学在治疗前预测局部进展期胃癌新辅助化疗疗效的价值。方法选取两家不同省份肿瘤医院从2013年6月~2020年11月接受新辅助化疗并行根治性手术的312例胃癌患者。根据术后病理组织学评估,将所有的患者分为新辅助化疗反应良好组和反应不良组。手动勾画每例患者图像病灶最大区域并提取2164个特征,经可重复性分析及支持向量机递归特征消除算法最终选出4个特征,利用逻辑回归模型构建影像组学标签。另外,通过多因素Logisitc回归分析患者的临床病理资料,预测新辅助治疗反应的价值。结果影像组学标签在训练集上预测胃癌新辅助化疗反应良好的ROC曲线下面积、敏感度、特异度、准确度分别为0.786(95%CI:0.679~0.894)、0.722、0.833、0.778,在验证集上分别为0.759(95%CI:0.656~0.863)、0.821、0.631、0.679。多因素Logistic回归显示临床病理学因素不是胃癌新辅助化疗疗效的独立预测因子。结论CT影像组学标签可作为预测胃癌新辅助化疗疗效的一种新型生物标记物,具有较好的预测效能。

Objective To investigate the value of computed tomography(CT)-based radiomics signature prediction of the neoadjuvant chemotherapy outcomes for locally advanced gastric cancer through multicenter study.Methods A total of 312 patients with gastric cancer who received neoadjuvant chemotherapy and radical surgery from June 2013 to November 2020 were collected,and they were hospitalized in two cancer hospitals from different provinces.According to the postoperative histopathological evaluation,all patients were divided into two groups,namely,good response group and none-good response group.The largest region of lesion was drawn manually and 2164 features were extracted.After repeatability analysis and the support vector machine recursive feature elimination algorithm,4 features were selected,and logical regression model was used to construct the radiomics signature.Besides,the value of clinicopathological data of patients was analyzed by Multivariate Logistic regression.Results The areas under the curve of ROC,sensitivity,specificity and accuracy were 0.786(95%CI:0.679~0.894),0.722,0.833 and 0.778 on the training set,respectively,and 0.759(95%CI:0.656~0.863),0.821,0.631 and 0.679 on the validation set,respectively.Multivariate Logistic Regression revealed that clinicopathological factors such as tumor histological differentiation were not independent predictors of neoadjuvant chemotherapy for gastric cancer.Conclusion CT-based radiomics signature can be used as a biomarker to predict the efficacy of neoadjuvant chemotherapy for locally advanced gastric cancer.