基于双能量CT影像组学模型术前预测进展期胃腺癌短径≥0.6cm淋巴结转移

Radiomics model based on dual energy CT for preoperativepredicting short axis ≥0.6 cm lymph node metastasis ofadvanced gastric adenocarcinoma

目的分析基于双能量CT影像组学模型术前预测进展期胃腺癌短径≥0.6 cm淋巴结转移(LNM)的价值。方法回顾性分析经手术切除的进展期胃腺癌患者,根据病理结果纳入36例pN3期114枚转移淋巴结(转移组)和26例pN0期65枚非转移淋巴结(非转移组),入组淋巴结短径均≥0.6 cm,将淋巴结分为训练集(n=125)和验证集(n=54)。对比组间原发肿瘤及淋巴结CT特征,采用广义估计方程(GEE)构建临床模型。提取静脉期融合图和碘图中的淋巴结影像组学特征,以组内相关系数(ICC)检验和Boruta算法筛选特征,构建影像组学模型,采用受试者工作特征(ROC)曲线和决策曲线分析(DCA)评价模型的诊断效能和临床收益。结果单因素及多因素GEE分析显示,原发肿瘤部位及最大径、淋巴结边缘及脂肪分数为LNM独立预测因素(P均<0.05),以之构建的临床模型预测训练集和验证集LNM的曲线下面积(AUC)分别为0.74和0.76。经ICC检验(ICC>0.8)及Boruta算法筛选,最终保留27个影像组学特征;以之建立的影像组学模型预测训练集和验证集LNM的AUC分别为0.99和0.98,均高于临床模型(P均<0.01),且临床收益更优。结论基于双能量CT影像组学模型术前预测进展期胃腺癌短径≥0.6 cm LNM具有较高价值。

Objective To explore the value of radiomics model based on dual-energy CT for preoperative predicting axis≥0.6 cm lymph node metastasis(LNM)of advanced gastric adenocarcinoma.Methods Data of patients with gastric adenocarcinoma who underwent surgical operation were retrospective analyzed.According to pathological findings,114 metastatic lymph nodes in 36 pN3 stage patients(transfer group)and 65 non-metastatic lymph nodes in 26 pN0 stage patients(non-transfer group)were enrolled,and the short axis of the enrolled lymph nodes were≥0.6 cm.The lymph nodes were divided into training set(n=125)and validation set(n=54).CT features of primary tumors and lymph nodes were compared between groups,and generalized estimating equation(GEE)was used to construct a clinical model.Radiomics features of lymph nodes were extracted from venous phase fusion images and iodine maps,while intra-class correlation coefficient(ICC)test and Boruta algorithm were used to filter features to construct a radiomics model.Receiver operating characteristic(ROC)curves were drawn,and decision curve analysis(DCA)was performed to evaluate the diagnostic efficacy and clinical benefit of models.Results Univariate and multivariate GEE analyses showed the location and the maximum diameter of primary tumor,as well as lymph node margin and fat fraction were all independent predictors of LNM(all P<0.05).The area under the curve(AUC)of clinical model constructed based on the above parameters was 0.74 and 0.76 in training set and validation set,respectively.Radiomics model was constructed based on 27 radiomics features filtered by ICC test(ICC>0.8)and Boruta algorithm,with AUC of 0.99 and 0.98 in training set and validation set,respectively,both higher than those of the clinical model(both P<0.01)and had better clinical benefit.Conclusion Radiomics model based on dual-energy CT had good value of preoperative prediction of axis≥0.6 cm LNM of advanced gastric adenocarcinoma.