| Objective:1.To investigate the value of clinical-CT model in predicting occult peritoneal metastasis in advanced gastric cancer.2.To explore the value of(2D,3D)radiomics models in predicting occult peritoneal metastasis in advanced gastric cancer.3.Establish and validate a combined model for preoperative prediction of occult peritoneal metastasis of advanced gastric cancer based on CT radiomics characteristics of primary tumors.Materials and methods:1.From January 2016 to December 2018,49 patients with advanced gastric cancer with occult peritoneal metastasis were detected during the operation,and49 patients with advanced gastric cancer without distant metastasis(including no peritoneal metastasis)were randomly selected,a total of 98 patients were included in the training set.From January 2019 to August 2020,15 patients with advanced gastric cancer with occult peritoneal metastasis were detected during the operation,and 15 patients with advanced gastric cancer without distant metastasis(including no peritoneal metastasis)were randomly selected,a total of 30 cases were included in the external validation set.2.Collected clinical and imaging data of patients,including: gender,age,preoperative biopsy pathology,degree of differentiation,Borrmann classification,cancer antigen 125(CA125),cancer antigen CA19-9(CA19-9),and the tumor TN staging on CT images,tumor size,the main part of the tumor,and mild ascites.Univariate and multivariate logistic regression analysis was used to find the independent predictors of occult peritoneal metastasis in gastric cancer,and a clinical-CT prediction model was established,and the model was visualized using a nomogram.Used receiver operating characteristic curve(ROC)area under the curve(AUC)and 95% confidence intervals(CI),as well as sensitivity and specificity to evaluate the diagnostic performance of the model.3.Used the Rad cloud platform to segment the tumor ROI(region of interest)on the 1.5-2.0 mm thick venous CT image.Two-dimensional ROI(2D)covers the largest cross-sectional area of the tumor.Three-dimensional ROI(3D)covers the entire tumor.The inter-reader reliability and intra-reader reproducibility of radiomics feature extraction were evaluated by calculating inter-and intra-class correlation coefficients(ICCs).The least absolute shrinkage selection operator algorithm(LASSO)was used to select radiomics features.Combined the selected features,used Support Vector Machine(SVM),Logistic Regression(LR),Multilayer Perceptron(MLP),Random Forest(RF)classifiers to construct 2D and 3D radiomics model respectively,and selected an optimal classifier model in each of the 2D and 3D radiomics model.4.Delong test was used to evaluate the difference between the(2D,3D)radiomics models in the training set and the external validation set of the ROC curve(AUC).Selected the best radiomics model as the radiomics signature(RS)and combined the clinical predictors to establish a combined model,and used a nomogram to visualize the combined model.Result:1.Multivariate logistic regression analysis showed that tumor size,mild ascites and CA125 were independent risk factors for peritoneal metastasis of gastric cancer.The AUC of the clinical-CT model for predicting occult PM of AGC in the training set and external validation set were 0.933(95% CI,0.886-0.981)and 0.853(95% CI,0.718-0.989),respectively.The sensitivity and specificity of this model for predicting occult PM in the training set were 0.837 and 0.959,respectively;in the external validation set,the sensitivity and specificity of predicting occult PM were 0.667 and 0.933,respectively.2.Of the 1409 radiomics features extracted from 2D and 3D ROIs,972 2D and 1055 3D radiomics features showed good inter-observer and intra-observer consistency(ICCs>0.75).For the 2D radiomics model,12 optimal features with non-zero coefficients were selected by LASSO,and 2D radiomics model was constructed using LR.The AUC of this model for predicting PM in the training set and external validation set were 0.841(95%CI,0.765-0.917),0.622(95%CI,0.411-0.833).The sensitivity and specificity of the 2D radiomics model for predicting occult PM in the training set were 0.857 and 0.735,respectively.In the external validation set,the sensitivity and specificity of the model for predicting occult PM were 0.800 and 0.467,respectively.For the 3D radiomics model,the same feature screening method(18 features retained)is used,and LR is also used to construct 3D radiomics model.The AUC of this model for predicting PM in the training set and external validation set was 0.885(95 % CI,0.851-0.948),0.676(95% CI,0.462-0.889).The sensitivity and specificity of the3 D radiomics model in predicting occult PM in the training set were 0.816 and0.816,respectively.In the external validation set,the sensitivity and specificity of the model in predicting occult PM were 0.600 and 0.860,respectively.3.The combined model contained these three indicators: mild ascites,CA125,3DRS.The predicted value of the calibration curve of this model had a good fit with the true value.The AUC of this model for predicting PM in the training set and external validation set were 0.940(95% CI,0.897-0.983)and0.796(95% CI,0.636-0.955),respectively.The sensitivity and specificity of this model for predicting occult PM in the training set were 0.755 and 1.000,respectively;in the external validation set,the sensitivity and specificity of predicting occult PM were 0.533 and 0.933,respectively.Conclusion:1.clinical-CT model predicted occult PM in AGCs with good diagnostic performance.The model was simple and interpretable,which is helpful to improve the preoperative diagnostic level of individual occult PM in patients with gastric cancer.2.(2D,3D)radiomics models had potential diagnostic value for predicting occult PM in patients with AGC.However,the radiomics models need to be further optimized and validated.3.Our study found that the clinical-CT model is better than radiomics models in predicting occult PM in patients with AGC.In addition,compared with the 2D radiomics model,the 3D radiomics model performs slightly better in predicting the occult PM of AGC. |
PDF Full Text Request