Value Of Radiomics Based On Transverse And Sagittal CE-T1WI In Pathological Grade Prediction Of Meningiomas

Background:Meningioma is the most common histopathological type of adult central nervous system(CNS)tumor,with an increasing incidence year by year.The 2021 World Health Organization(WHO)classification of CNS tumors(5th edition)retains the WHO grades of meningioma.The new classification has not substantially changed the management of meningioma,and the WHO grades still guide its treatment,which affects patients’ prognosis to a certain extent.Objective:To explore the value of radiomics based on transverse and sagittal contrastenhanced T1-weighted imaging(CE-T1WI)in predicting the pathological grading of meningiomas.Methods and materials:A retrospective analysis of MRI images of 945 patients with meningiomas,confirmed by surgical pathology and meeting inclusion criteria at Guangdong Sanjiu Brain Hospital between January 2017 and December 2021,was performed,including 811 patients with WHO grade 1 and 134 patients with WHO grade 2/3.First,univariate,correlation,and multivariate analyses were employed to study the relationship between clinical features,imaging features and pathological grading of meningiomas.A logistic regression clinical model was constructed and its efficacy was evaluated.Next,the tumor regions of interest(ROI)in transverse and sagittal CE-T1WI were manually delineated using the ITK-SNAP software,and the data were randomly divided into a training set(n=660)and a test set(n=285)at a 7:3 ratio.Radiomic features were extracted using the Pyradiomics software,and feature selection was performed through variance selection,Spearman rank correlation coefficient,Pearson correlation coefficient,T-test,and recursive feature elimination(RFE)algorithm based on logistic regression(LR)classifier.Support vector machines(SVM),adaptive boosting(Adaboost),gradient boosting(Gradboost),and random forest classifier(RFC)were used to construct a separate a transverse model and a combined transverse and sagittal model.The repeatability of features was evaluated using intraclass and interclass correlation coefficients(ICC).The diagnostic performance of these models was assessed using receiver operating characteristic(ROC)curves and area under the curve(AUC).The performance of different models was compared by calculating accuracy,precision,recall score,F1 score,sensitivity,and specificity.Results:Multivariate analysis showed that the maximum tumor diameter and heterogeneous enhancement were correlated with high-grade meningiomas.The established logistic regression clinical model had moderate efficacy in predicting meningioma pathological grades(AUC=0.68),with sensitivity and specificity of 0.799 and 0.493,respectively.Eighteen and fifteen radiomic features were selected from the training set(n=660)for the separate transverse model and the combined transverse and sagittal model,respectively.The separate transverse model based on SVM_linear had average efficiency(AUCtest=0.587),while the combined transverse and sagittal model based on Gradboost had fair efficiency(AUCtest=0.707),with sensitivity and specificity of 0.704 and 0.681,respectively.Conclusion:The radiomics method based on axial and sagittal CE-T1WI is helpful for preoperative pathological grading prediction of meningiomas to a certain extent,and can provide a basis for the formulation of clinical treatment plans.