Feasibility Study Of Predicting Pathological Grade Of Meningioma By MRI-based Radiomics

Objective:To explore the value of conventional MRI-based radiomics in predicting the pathological grade of meningioma.Methods:The MRI images of 137 patients with meningiomas confirmed by pathology were analyzed retrospectively.There were 99 cases of low grade meningioma(WHO I grade)and 38 cases of high grade meningioma(WHO Ⅱ grade).According to the proportion of 7:3,they were divided into training group(n=95)and test group(n=42).The MRI signs were evaluated,including the number of hemibrain cases,location,shape,vascular flow emptying,peritumoral edema,homogeneity of enhancement,degree of enhancement,meningeal tail sign,adjacent tissue invasion and growth across the midline.The age differences between the two groups were compared by independent sample t-test,and the differences of gender and image features between the two groups were compared by chi-square test or Fisher exact test.Five feature groups,including Intensity_Histogram,Gray_Level_Cooccurence_Matrix,Gray_Level_Run_Length_Matrix,Neighbor_Intensity_Difference_Matrix and Intensity_Direct,were manually extracted from ROI,drawing with IBEX software.736 radiomics features were extracted from each MRI sequence.Using the intra-group correlation coefficient(ICC)to evaluate the consistency of the researcher itself,the one-way analysis of variance was used to select the parameters with statistical significant differences,and then the least absolute shrinkage and selection operator(LASSO)and 10-fold cross-validation were used to reduce the dimension.Finally,the radiomics features obtained by dimensionality reduction are incorporated into four classification learners included logical regression(LR),support vector machine(S VM),random forest(RF)and adaptive boosting(AdaBoost).Four radiomics models are constructed,and the model with the best classification efficiency is selected.And then an image feature model is constructed by using the statistically significant image features of analysis of variance.The classification learner used in this model is consistent with that used in the previous optimal radiomics model.Finally,a combined diagnosis model is constructed by combining MRI image features and radiomics features,and the classification learner is the same as above.Draw the receiver operator characteristic curve(ROC)to evaluate the sensitivity,specificity and accuracy of the model prediction,and use the area under the ROC curve(AUC)to evaluate the prediction efficiency of the model.Results:There was no significant difference in the general clinical data of the patients.In the training group and the test group,only vascular flow emptying and homogeneity of enhancement were statistically significant in grade Ⅰ and grade Ⅱmeningiomas(p<0.05).Thirteen radiomics features were selected after dimensionality reduction.Among the four models of LR,SVM,RF and AdaBoost constructed by radiomics features,The sensitivity was 66.7%,the specificity was 86.7%,the accuracy was 76.3%,and the AUC value was 0.811 of the LR model.The sensitivity was 15.0%,the specificity was 53.3%,the accuracy was 64.3%,the AUC value was 0.606 of the SVM model.The sensitivity was 75.0%,the specificity was 76.7%,the accuracy was 78.5%,and the AUC value was 0.807 of the RF model.The sensitivity,specificity,accuracy and AUC value of AdaBoost model were 66.7%,76.7%,71.4%and 0.736 respectively.LR model has the highest AUC value.Image feature model and combined diagnosis model are constructed based on LR classification learner.The effectiveness of the image feature model,the radiomics model and the combined diagnosis model were evaluated respectively.The ROC curve showed that the sensitivity was 66.7%,the specificity was 80.0%,the accuracy was 71.4%,and the AUC value was 0.800 of the image feature model;the sensitivity,specificity,accuracy and AUC value of the radiomics model were 66.7%,86.7%,76.2%and 0.811,respectively.The sensitivity was 100.0%,the specificity was 66.7%,the accuracy was 78.6%,and the AUC value was increased to 0.856 of the combined diagnosis modelConclusion:1.In the features of MRI images,whether there is vascular flow emptying in the tumor and whether the enhancement is uniform or not have a certain diagnostic value for the grading of meningiomas.2.The radiomics model based on conventional T2WI and enhanced T1WI has important clinical application value in the grading diagnosis of meningioma,and the radiomics method has a good application prospect.3.The diagnostic efficiency of the combined diagnosis model is higher than that of the radiomics model and the image feature model.