Differentiation Of High-grade Gliomas And Solitary Brain Metastases Based On Different Radiomics Models Of Conventional Magnetic Resonance Imaging

Objective:To explore the combination of T2 WI and T1-weighted contrast-enhanced scan(T1/C)images with different machine learning algorithms and establish corresponding imaging omics models,and explore the imaging omics models to identify high-grade gliomas(High-grade gliomas,HGG)before surgery.)And the value of solitary brain metastasis(SBM).Materials and methods:Collect the menstrual cases or follow-up confirmed patients with HGG(51 cases)and SBM(45 cases)in our hospital from January 2016 to July 2020.All patients underwent routine MRI(including T2 WI and T1/C)examinations before the operation,and passed ITK-SNAP software,delineate the region of interest(ROI)layer by layer on the T2 WI and T1/C axis images respectively,obtain the volume of interest(VOI),and perform imaging omics Feature extraction,all cases are divided into training group and test group according to 70%:30%.The training group is used for feature screening and the establishment of imaging omics model.Feature screening is completed by ttest and LASSO.Random forest is selected for the data after feature screening(RF),Logistics regression(LR)and support vector machine(SVM)are three commonly used machine learning algorithms and establish the corresponding imaging omics model;the test group is used to verify the established model and draw the ROC curve.The result is expressed as accuracy,Sensitivity,specificity and AUC.Results:There was no significant difference in age and gender composition between HGG group and SBM group(P>0.05).Among the models based on the features extracted from a single sequence,the three models based on T2 WI had moderate diagnostic performance with little difference.Among the imaging omics models based on T1/C,the RF model had the highest diagnostic performance,with an AUC of 0.90;Models based on different single sequences,the T1/C-based RF and SVM models have significantly improved diagnostic performance compared with the T2 WI models,and the LR model has little change in diagnostic performance;combined with all the features extracted from T2 WI and T1/C sequences to build 3 All models showed high diagnostic performance,and the SVM model had the highest diagnostic performance,with an AUC of 0.90.Conclusions:The imaging omics model is of high value in the identification of high-grade gliomas and solitary brain metastases.Different magnetic resonance sequences and different machine learning algorithms have an impact on the diagnostic performance of the model,which comes from multiple sequences and combined features The imaging features may provide more comprehensive information for the differential diagnosis of tumors.