The Value Of CT-based Radiomics In Preoperative Prediction Of Lymph Node Metastasis And Degree Of Pathological Differentiation In Gastric Cancer

Part Ⅰ: The value of texture analysis based on spectral CT in predicting preoperative lymph node metastasis of gastric cancerObjective: To study the value of spectral CT texture analysis in predicting lymph node metastasis in patients with gastric cancer before operation.Methods: 80 patients with gastric cancer confirmed by surgical resection and pathology were analyzed retrospectively（57 cases in training group and 23 cases in verification group）.The special texture analysis software AK was used to segment the lesions and extract the imaging features on the preoperative energy spectrum CT 70 kev venous phase cross-sectional images.The Mann-Whitney U test was used to analyze the features between the two groups,and the p < 0.05 features were preserved.The discriminative features were further found by single factor logical regression analysis.Then the minimum redundancy maximum correlation method was used to eliminate the10 features with the highest correlation with the label.Stepwise multiple logical regression was used to construct the prediction model and the final model.The performance of the model is evaluated by ROC analysis.Results: in terms of texture features,the 10 radiology-related features selected by MRMR had better discriminative ability for training group and verification group（AUC > 0.64）,and the AUC of multivariate logical regression prediction model was0.79（0.69-0.89）.Conclusion: texture analysis based on energy spectrum CT is expected to be a noninvasive tool for predicting preoperative lymph node metastasis in patients with gastric cancer.Part Ⅱ: A CT-based radiomics nomogram for preoperative prediction of differentiation degree in advanced gastric cancerPurpose: To establish and validate the value of the CT-based radiomics nomogram in the preoperative prediction of differentiation degree in advanced gastric cancer.Materials and Methods: This retrospective study enrolled 260 patients with surgicalpathologically confirmed advanced gastric cancer from two centers（165 patients in the training cohort,55 patients in the testing cohort,and 40 patients in the external testing cohort）.The region of interests（ROIs）were manually depicted on arterial phase and venous phase CT images,respectively.The least absolute shrinkage,selection operator regression and iterative screening features were applied for feature selection and radiomics signature construction.An external testing cohort was used to verify the stability of radiomics signature model.Finally,nomogram was built integrating the radiomics signature and significant clinical-radiological predictors using logistic regression,respectively.The utility of the prepared model was evaluated using receiver operating characteristic curves,calibration curve,and decision curve analysis.Results: Compared to the venous phase-based model,the arterial phase-based model showed better predictive performance.External testing cohort verified the stability of the arterial phase-based model（AUC = 0.789）.Through the Logistic Regression classifier,the nomogram integrating the arterial phase-based radiomics signature and clinical-radiological parameters showed the best prediction performance in the training and test cohorts with AUCs of 0.877（95% CI,0.825 to 0.930）and 0.854（95% CI,0.756 to 0.953）,respectively.The efficacy of the radiomics nomogram was higher than that of the combined model.The ROC curves,calibration curve,and decision curve analysis demonstrated the clinical usefulness of the developed nomogram.Conclusions: The CT-based radiomics nomogram could be used as a noninvasive tool for the preoperative prediction of differentiation degree in patients with advanced gastric cancer.