The Prediction Of Meningioma Grade Using Multimodal Magnetic Resonance Imaging-Based Radiomics

Objective: To evaluate the role of radiomic model based on conventional magnetic resonance imaging(c MRI),apparent diffusion coefficient(ADC)map or sensitivity weighted imaging(SWI)in predicting the meningioma grade.And to investigate the prediction performance of radiomic models based on multimodal MRI using the combination of c MRI,ADC map,SWI in order to search the optimal model and improve the accuracy of preoperative prediction of meningioma grade.Materials and Methods: Two hundred and twenty-nine low-grade(Grade I)and87 high-grade(Grade II/III)patients with pathologically diagnosed meningiomas from August 2009 to May 2019 were enrolled.All patients underwent c MRI,diffusion weighted imaging(DWI)and SWI.The MRI features including tumor location,necrosis or cyst degeneration,hemorrhage and calcification were analyzed.Radiomic features from c MRI,ADC map and SWI were extracted based on the volume of entire tumor.Classification performance was evaluated by a nested leave-one-out cross-validation(LOOCV),combining the least absolute shrinkage and selection operator(LASSO)feature selection and random forest(RF)classifier that was trained(1)without subsampling,and(2)with the synthetic minority over-sampling technique(SMOTE).The prediction performance of different radiomic models(c MRI,ADC,SWI,c MRI+ADC,c MRI+SWI,ADC+SWI and c MRI+ADC+SWI models)were assessed using receiver operating characteristic(ROC)curve and area under the curve(AUC)of them were compared using Delong’s test.Results: Without subsampling,the AUCs of c MRI,ADC and SWI models were0.79,0.75 and 0.71,respectively.The AUC showed no significant difference between c MRI+ADC model and c MRI model(0.82 vs 0.79,P>0.05).The ADC+SWI model showed significantly lower AUC than c MRI model(0.70 vs 0.79,P<0.05).The c MRI+SWI model yielded the higher AUC than c MRI model(0.89 vs 0.79,P<0.05),while there were no significant difference in AUC between c MRI+SWI model and c MRI+ADC+SWI model(0.89 vs 0.89,P>0.05).In terms of using SMOTE approach,the AUC of c MRI、ADC and SWI models were 0.77,0.76 and 0.68,respectively.The AUC showed no significant difference between c MRI+ADC model and c MRI model(0.80 vs 0.77,P>0.05).The ADC+SWI model showed significantly lower AUC than c MRI model(0.67 vs 0.77,P<0.05).The c MRI+SWI model yielded the higher AUC than c MRI model(0.87 vs 0.77,P<0.05),while there were no significant difference in AUC between c MRI+SWI model and c MRI+ADC+SWI model(0.87 vs 0.86,P>0.05).Conclusions:(1)c MRI,ADC and SWI radiomic model may be helpful for the prediction of meningioma grade;(2)Radiomic model based on c MRI and SWI yielded the best prediction performance,and radiomic model based on c MRI,ADC map and SWI can’t further improve the prediction performance.