Objective:To explore the value of preoperative prediction of lymph node metastasis(LNM)in gastric cancer based on CT radiomics nomogram.Methods:Clinical and imaging data of 208 patients with gastric cancer were retrospectively collected.All patients were divided into a training cohort(147 cases)and a validation cohort(61 cases)at a 7∶3 ratio,and the patients in each cohorts were divided into the LNM and non-LNM groups.The lesions'ROIs were manually delineated on plain scan,arterial phase,venous phase and delayed phase images using ITK-SNAP software,and the lesions'VOIs were obtained.Then,the texture features of VOIs were extracted using FAE v.3.7.7z software.The minimal-Redundancy-Maximal-Relevance(mRMR)and LASSO methods were used to screen the radiomics features of plain scan,arterial phase,venous phase,delayed phase and the combined multiple phases for predicting LNM in gastric cancer,and the radiomics models were established and the the rad-scores were calculated.Multivariate logistic regression analysis was used for the independent predictors.The combined clinical and radiomics model was constructed,and the nomogram was drawn and used to analyze the prediction probability.The predictive efficiency,calibration and clinical application value of these models were evaluated using ROC curve,Hosmer-Lemeshow test and decision curve analysis(DCA).Results:The AUCs of the radiomics models of plain scan,arterial phase,venous phase,delayed phase and the combined multiple phases in prediction LNM in the training cohort were 0.80,0.76,0.81,0.76 and 0.79,respectively,with the venous phase radiomics model having the highest predictive efficiency.The AUCs of clinical model in the training cohort and validation cohort were 0.79 and 0.81,and those of the combined clinical and radiomics model were 0.86 and 0.82,which were significantly superior to clinical model,and the venous rad-score was an independent risk factor(OR=2.45,P<0.05).DCA showed the net return of the combined clinical and radiomics model was higher than that of the clinical model and the venous radiomics model with the risk of 26%～88%.Conclusions:LNM,tumor thickness showed on CT and venous rad-score are independent risk factors for predicting LNM in gastric cancer,and the diagnostic efficiency of the combined three factors is higher.The nomogram established based on CT features at venous phase has a high predictive ability for LNM of gastric cancer.
国家科技期刊平台
NSTL
万方数据
