Objective:To investigate the predictive value of CT-based radiomics in the prediction of normal-sized lymph node metastasis at different sites of gastric cancer.Methods:A retrospective analysis was performed for 303 normal-sized lymph nodes in 233 patients with gastric cancer that met the inclusion criteria,and the lymph nodes were divided into two zones,Zone 1(the first station lymphnode,175 lymph nodes)and Zone 2(the second and third station lymphnode,128 lymph nodes),and the lymph nodes of the two zones were divided into the training and validation cohorts at a ratio of 7∶3,respectively.The clinical-imaging combination model was constructed for the lymph nodes in the two zones,by screening the clinical independent risk factors and radiomics characteristics,and using four different classifiers,decision tree,LinearSVC,SVM and logistic regression.The performance of the combination model was evaluated by AUC,and the clinical value of the model was analyzed by DCA.Results:Four clinical-imaging combination models were established in the two zones,and the SVM model of Zone 1 showed the best discrimination performance with the AUCs of 0.960(95%CI 0.925～0.995)and 0.731(95%CI 0.556～0.905)in the training and validation cohorts,respectively.The SVM model of Zone 2 also showed the best discrimination performance with the AUCs of 0.998(95%CI 0.993～1.000)and 0.959(95%CI 0.924～0.994)in the training and validation cohorts,respectively.DCA showed that the application of the SVM combination models of the two zones in clinical decision-making achieved great benefits.Conclusion:The clinical-imaging combination model based on CT radiomics can predict the normal-sized lymph node metastasis at different sites of gastric cancer before surgery.
国家科技期刊平台
NSTL
万方数据
