The Predictive Value Of Contrast-enhanced CT Based Radiomics In Preoperative CT4 Stage Of Advanced Gastric Cancer

Objective: Only 42-66% of cT4 b patients have the same pathological stage,which suggests that better diagnostic methods are needed for the accurate identification of T4 b stage.The purpose of this study was to explore the possibility of the radiomics method based on contrast-enhanced computed tomography(CE-CT)in differentiating gastric cancer patients with preoperative cT4 stage into p T4 b stage and no-p T4 b stage.Methods: From January 2008 to December 2021,the date of 691 gastric cancer patients(460 in the training set,231 in the validation set)who underwent preoperative CE-CT examination and were staged as cT4 in the Affiliated Hospital of Qingdao University were retrospectively collected.According to the pathological stage,the patients were divided into p T4 b stage and no-p T4 b stage.In the training set,the clinical characteristics and the radiomics features extracted from the venous-phase CE-CT images were analyzed to construct the clinical model,radiomics model and radiomics combined clinical nomogram.Two kinds of radiomics feature selection methods were utilized to dimensionality reduction:The LASSO and m RMR algorithm.The SVM algorithm is utilized to as machine learning classifiers.The model performance was quantized by the calibration curve,DCA curve and AUC.Results:A total of 691 patients were included in this study.According to preoperative CECT evaluation,317 patients were cT4 a stage and 374 patients were cT4 b stage.According to the pathological results,332 patients were classified as p T4 b stage and 359 patients were classified as no-p T4 b stage.Borrmann classification,cT stage and Rad-score were utilized to construct clinical combined radiomics nomogram.The AUC of the clinical model was0.887(95% confidence interval(95% CI): 0.843-0.912)and 0.853(95% CI: 0.821-0.923)in the training and validation sets.The AUC for the radiomics model were 0.866(95% CI:0.832-0.899)and 0.781(95% CI: 0.711-0.846)in the training and validation sets respectively.The AUC of the nomogram were 0.940(95% CI: 0.912-0.983)and 0.876(95%CI: 0.824-0.931)in the training and validation sets respectively.Considering the performance of the radiomics model in the validation set was inferior to the clinical model,in order to further clarify the pros and cons of the radiomics model and the cT stage subjectively assessed by the radiologist.Univariate logistic regression analysis was performed on cT stage,the AUC of cT stage were 0.860(95% CI: 0.826-0.872)and 0.805(95% CI: 0.746-0.862)in the training and validation sets respectively.Delong test was performed on the AUC of nomogram,clinical model,radiomics model,and cT stage characteristics.In the validation set,the nomogram achieved the highest AUC and the clinical model achieved the second highest AUC,which got statistical difference from the other models.As for cT staging and radiomics model,the AUC of cT staging was higher than that of radiomics model,but no statistical difference was found.Therefore,the predictive ability of the models was compared: nomogram > clinical model > cT stage≥ radiomics model.The calibration curve of the nomogram shows a good calibration,indicating a good agreement between predictions and actual results.This study evaluates the DCA curve of the model’s the clinical application value.All four models achieved satisfactory clinical value.In the validation set,using radiomics model,cT stage,clinical model,nomogram to predict p T4 b stage was better than treating all patients or not treating all patients,which will achieve more clinical positive net benefit.Within a wide range of risk thresholds,the comparison of clinical positive net benefit: nomogram > clinical model >cT stage ≈ radiomics model.Conclusion: The performance of the radiomics model based on venous phase CE-CT images was not superior to clinical models and subjective assessment of cT stage,indicating the radiomics method of this study has certain limitations in distinguishing patients with p T4 b stage.However,the clinical combined radiomics nomogram achieved the best predictive performance and clinical benefit,indicating that the clinical combined radiomics method can improve the model performance,which has a certain reference value for the selection of treatment methods for patients with cT4 stage gastric cancer.