Application Of Radiomics Based On Parotid Gland ZOOMit DWI In Sj(?)gren’s Syndrome

Objective:To explore the application value of radiomics model based on parotid gland ZOOMit DWI imaging in predicting the diagnosis of primary Sj(?)gren’s syndrome.Methods:In this study,38 patients with clinically diagnosed primary Sj(?)gren’s syndrome were selected as the experimental group,and 38 healthy volunteers of similar gender,age and eligible non-p SS volunteers were also matched as the control group.All patients’parotid glands had conventional scanning and ZOOMit DWI scanning,while on the same Siemens Prisma 3.0T MRI.Based on ZOOMit DWI images(b=800 s/mm~2),the bilateral parotid glands were drawing by manual and the image features were extracted,and the model was constructed by ten-fold cross validation.Select logistic regression,support vector machine,random forests and extreme gradient boosting classifiers training model,get the corresponding classifier training model and verify the prediction model performance.The receiver operating curve was used to evaluate the four models performance indexes:area under the ROC curve,specificity,sensitivity,accuracy,F1 score,and Delong test was used to test whether the AUC value difference between the pairwise comparison models.Results:1.LR model training set:AUC valve of 0.9307,accuracy of 0.8553,sensitivity of 0.9474,specificity of 0.7632,F1 score of 0.8675,validation set:AUC valve of 0.9044,accuracy of 0.8026,sensitivity of 0.8421,specificity of 0.7632,and F1 score of 0.8101.2.The SVM model training set:AUC valve of 0.9335,accuracy of 0.8553,sensitivity of 0.9474,specificity of 0.7632,F1 score of 0.8675,the validation set:AUC valve of0.9044,accuracy of 0.8289,sensitivity of 0.8947,specificity of 0.7632,and F1 score of 0.8395.3.The RF model training set:AUC valve of 0.9841,accuracy of 0.9079,sensitivity of0.9474,specificity of 0.8684,F1 score of 0.9114;the validation:AUC valve of 0.883,accuracy of 0.8026,sensitivity of 0.8421,specificity of 0.7632,and F1 score of0.8101.4.The XGBoost model training set:AUC valve of value 0.9273,accuracy of 0.8684,sensitivity of 0.9737,specificity of 0.7632,F1 score of 0.881,validation set:AUC valve of 0.903,accuracy of 0.8158,sensitivity of 0.8684,specificity of 0.7632,and F1score of 0.825.5.The Delong test results showed significant differences between the RF model with LR,SVM,and XGBoost(P=0.012,0.035,0.035,0.013),and the highest AUC value of the RF model was 0.9841;none of the remaining three models significantly(P>0.05).None of the models were significant in the four groups in the validation set(P>0.05).Conclusion:1.The training set and validation set of LR,SVM and XGBoost prediction model all have high AUC value,which achieves a good diagnostic efficacy for the model of predicting primary Sj(?)gren’s syndrome.2.The prediction model constructed by RF had overfit conditions,and the overall model had poor diagnostic performance.