Diagnosis Efficacy Of Serosa Invasion Using A Radiomics Model Focusing On Gastric Adenocarcinoma And Peritumoral Radiomics Features In Dual Source CT Dual Energy Imaging

Objective: To establish a radiomics model based on gastric adenocarcinoma and peritumoral radiomics features derived from 120-k V equivalent mix images and iodine map(IM)of dual source CT(DSCT)dual energy imaging in the portal venous phase,and evaluate the diagnostic efficacy of it for detecting serosal invasion.Methods: 231 patients with postoperative pathologically confirmed gastric adenocarcinoma were retrospectively enrolled and randomly divided into three cohorts: training(n=132),testing(n=58),and independent validation(n=41)cohorts.Traditional features(including clinical and semantic features)were analysed by two radiologists using a dichotomous classification method.A rectangular region of interest(ROI)was manually delineated from the mucosal surface to 5 mm outside the serosal surface perpendicular to the gastric wall in the direction perpendicular to the gastric wall on the deepest layer of tumor infiltration using Radiomics software.The radiomics features were extracted from the ROI of the 120-k V equivalent imaging and IM.The most significant features were selected by intraclass correlation consistency analysis and Boruta algorithm.The random forest algorithm was used for radiomics model construction.Univariate and multivariate logistic regression was used to construct the clinical model and combined clinical-radiomics model.The diagnostic efficacies of the models were evaluated using receiver operating characteristic(ROC)analysis and the area under the curve(AUC)of each model was calculated,and the AUC values of each model were compared using the De Long’s test.The calibration curves were used to evaluate the calibration performance of the combined model.The clinical effectiveness of the three models were compared using decision curve analysis(DCA).Results: A clinical model based on one traditional feature(increased fat density of serosal surface),a radiomics model based on nine radiomics features,and a combined clinical-radiomics model based on traditional features and Rad-score were constructed.A statistically significant difference was noted between the Rad-scores of patients with serosal invasion and without serosal invasion in the training,testing,and independent validation sets(all P < 0.001).Both the radiomics model and the combined model showed a high efficacy,with AUC of 0.90,0.90,and 0.85 for radiomics model and 0.93,0.93,and 0.89 for combined model,and outperformed the clinical model in the training,testing and independent validation sets(AUC: 0.76,0.76 and 0.81).Conclusions: The radiomics model and combined clinical-radiomics model constructed based on gastric adenocarcinoma and peritumoral radiomics features derived from DSCT dual energy imaging showed high diagnostic efficacy for serosal invasion in gastric adenocarcinoma,which can provide a basis for decision making in gastric adenocarcinoma treatment.