Preoperative Discrimination Between Solitary Brain Metastasis And Glioblastoma Multiforme By Using Radiomics Analysis

Purpose: To explore the value of radiomics analysis to distinguish glioblastoma(GBM)from solitary brain metastasis(sMET)at conventional MRI.Materials and Methods: Total 177 patients with pathologically confirmed GBM(n = 81,49 males and 32 females,average age: 49.32±16.55)and sMET(n = 96,55 males and 41 females,average age: 57.18±10.97)from 2007 to 2019 were retrospectively included.Patients were randomly divided into the training group(n=125,GBM=49,sMET=76)and the validation group(n=52,GBM=32,sMET=20)according to the time.All patients completed the examination of T2WI?FLAIR and T1WI-CE.The ITK-snap image application was used to manually sketch 3D ROI layer by layer.The ROI was delineated by using the image of T1WI-CE,and then the ROI of T2 WI and FLAIR were depicted using the T1WI-CE reference.The radiomics metrics were be extracted from standardized images and the intraclass correlation coefficient(ICC)was first carried out for repeatability test.And then the LASSO regression method was applied to further filtrate the features after removing the highly correlation feature.The optimal classifier model was established according to the residuary characteristics.The receiver operating curve(ROC)was drawn to assess the specificity,sensitivity,and the area under the curve(AUC)of each model to determine the optimal model.Results: Using T2 WI,FLAIR,T1WI-CE and the mixture sequences of the above 3 scans,the 21 features were selected by Lasso regression analysis,including 8 morphological features,13 texture features(GLCM 5,GLDM 4,GLSZM3,GLRLM1),3 features of T2 WI,7 features of FLAIR,5 features of TIWI-CE,and 6 features of the mixture sequences.In the training and validation data sets,there was a significant difference between GBM and metastatic tumors in the imaging Radiomics score(Rad-score)based on T2 WI,FLAIR,T1WI-CE and the combination of the three sequences(P < 0.001).All Rad-score based on these sequences are showed excellent identification performance.In the training group,the AUC of T2 WI reached 0.915(0.866,0.963),and the performance was about 0.848.Based on FLAIR,AUC in the training group reached 0.957(0.927,0.988),and the diagnostic level was about 0.880.And the characteristic AUC of the training group based on T1WI-CE reached 0.950(0.916,0.983),and the diagnostic performance was about 0.888.The AUC of the mixture series was 0.947(0.911,0.983)and its diagnosis ability was about 0.896.Conclusion: The radiomics combine with MRI images can be used to distinguish GBM and sMET,which is helpful for the clinicians to make their decisions.