| Purpose1.To extract and quantify the PET/CT image feature,establish radiomics signiture closely related to vascular invasion in gastric cancer,and then evaluate the diagnostic efficacy of radiomics signiture in gastric cancer.2.To obtaine the independent predictors of vascular invasion in gastric cancer by multivariate logistic regression analysis of clinical feature and radiomics signature,and to evaluate the possibility and diagnostic efficacy of vascular invasion in gastric cancer preoperatively by constituting the PET/CT nomograms with radiomics signature and clinical feature.Methods92 patients with gastric adenocarcinoma confirmed by surgery and pathology in our hospital were retrospectively collected from January 2011 to November 2017.All patients were up to the inclusion and exclusion criteria and PET/CT examination were performed preoperatively.The prediction model was divided into training set with 62 patients and test set with 30 patients by computer random numbers.The clinical characteristics of each patient(patient age,gender,tumor thickness,SUVmax,tumor site,Borrmann typing,T staging,etc.)were evaluated and recorded by two attending PET/CT doctors.The ROI of indeterminate shape was placed on the largest cross section of PET and CT images,and the two-dimensional outline of the overall tumor was manually sketched.PET/CT characteristics were extracted by the analysis of machine learning method,and the Mann-Whitney U test was performed to retain the 36 significant features.The first 20 features with good repeatability were selected by mRMR method,which were used to remove the lengthy radiomics featurs.Then the 8 radiomics features,optimally related to vascular invasion in gastric cancer,were obtained through LASSO analysis,and the radiomics signature was established.The independent predictors of vascular invasion in gastric cancer were obtained by multivariate logistic regression analysis with radiomics signatures and clinical characteristics.Then a nomogram,combined with the clinical characteristics and the PET/CT radiomics signature was constructed to evaluate preoperatively the vascular invasion in gastric cancer by the regression coefficient,and finally the diagnostic efficacy of the radiomics model was evaluated by ROC and AUC.Results1.The differences in age and tumor site of training concentration were statistically significant(P = 0.045 and 0.017,respectively);There were statistically significant differences in tumor site and tumor thickness(P = 0.021 and 0.008,respectively).There were statistically significant differences in tumor site between the two groups(P<0.05).2.Three major categories and 1177 texture features were extracted from PET/CT images.After removing poor repeatability and redundant features with statistical methods,a total of 8 radiomics features with high dimensions and high repeatability were obtained to establish the radiomics signiture of vascular invasion in gastric cancer.3.The constructed PET/CT radiomics signiture had a high prediction of vascular invasion in gastric cancer in both the training set and the test set,with an AUC of 0.762 and 0.756,respectively.4.Tumor site and PET/CT histological labels were independent factors for preoperative prediction of vascular invasion in gastric cancer by multivariate regression analysis of clinical features and radiomics signiture with regression coefficients of 0.962 and 0.960,respectively.5.The nomogram was constructed by combining the radiomics signiture with tumor site.When the optimal diagnostic threshold was 0.70,the training set and the test set showed high diagnostic efficiency,with an AUC of 0.778 and 0.801,respectively.Conclusion1.This study was established firstly that radiomics nomogram based on PET/CT can predict preoperatively vascular invasion in gastric cancer.2.Tumor site and PET/CT radiomics signiture were independent predictors of vascular invasion in gastric cancer.3.In this study,the PET/CT radiomics signiture and clinical signiture(tumor site)were incorporated into the establishment of nomogram model,which was further optimized the diagnostic efficacy of the radiomics model for preoperative vascular invasion of gastric cancer,and provided more personalized decisions for clinical diagnosis. |
PDF Full Text Request