Prediction Of The Risk Categorization In Anterior Mediastinal Lesions Using CT-based Radiomics

PurposeA retrospective study was carried out to predict the risk categorization of anterior mediastinal lesions using CT-based radiomics,so as to distinguish the high-risk lesions from the low-risk lesions.This is helpful to better guide the clinical prognosis and treatment.Material and MethodsThis is a retrospective study.A total of 298 patients who had pathologically confirmed anterior mediastinal lesions were collected From February 2013 to March 2018.All Patients underwent CT scans before treatment,including 130 unenhanced computed tomography(UECT)and 168 contrast-enhanced computed tomography(CECT)scans.The lesions were segmented the lesions,and 1029 radiomics features were extracted.Then the Least absolute shrinkage and selection operator(LASSO)algorithm method was used to reduce the dimension of the features and select the best feature set.Finally,machine learning method of Logistic Regression(LR)was used to build the radiomics model.The prediction efficiency of model was evaluated by the receiver operating characteristic curve(ROC).ResultsAmong the 1029 image features in this study,12 best features were selected from the UECT and CECT scan images,respectively.Based on these features,he radiomics model was established.In the training set of UECT images,the AUC was 0.84(95% CI: 0.76-0.92,sensitivity was 0.71,specificity was 0.74);In the validation set of UECT images,the AUC was 0.77(95% CI: 0.63-0.92,sensitivity was 0.74,specificity was 0.73).In the training set of CECT images,the AUC was 0.74(95% CI: 0.63-0.81,sensitivity was 0.67,specificity was 0.65);In the validation set of CECT images,the AUC was 0.73(95% CI: 0.60-0.87,sensitivity was 0.63,specificity was 0.67).ConclusionRadiolomics models based on UECT and CECT can predict the risk categorization of anterior mediastinal lesions and distinguish high-risk lesions from low-risk lesions,and the radiomics model based on UECT scans have better prediction performance than CECT scans.