Prediction Pathological Grade And Ki-67 Proliferation Index Of Meningiomas Based On MRI Radiomics

Objective: The purpose of this study is to explore the predictive value of contrast-enhanced MRI radiomics in the pathological grade and expression level of Ki-67 in meningiomas.Materials and methods: This study retrospectively included 151 patients with pathologically confirmed meningiomas who were treated in our hospital from January 2016 to December 2020,of which 31 were high-grade(WHO grade Ⅱ and Ⅲ)and 120 were low-grade(WHO grade Ⅰ);The proliferation index Ki-67>4% was 45 cases,and Ki-67≤4% was 106 cases.According to the ratio of 7:3,all patients were randomly divided into training set and validation set.All patients underwent enhanced MRI scan before the operation.The patients’ MRI data were uploaded to Huiyi Huiying radiomics cloud platform,and the volume of interest(VOI)was outlined on magnetic resonance contrast-enhanced T1 weighted image(CE-T1WI).The cloud platform was used to extract and screen radiomics features,and the optimal feature subset was selected.Support vector machine and logistic regression machine learning were used to establish models to predict the pathological grade and Ki-67 expression level of meningiomas in the training set,and the models were verified in the validation set.Receiver operating characteristic curve(ROC)was drawn.Area under curve(AUC),sensitivity and specificity were used to evaluate the predictive effectiveness of the models.We assess conventional MRI imaging features including: the maximum diameter of the tumor,tumor location,tumor shape,peritumoral edema,meningeal tail sign,tumor enhancement degree and enhancement mode.Univariate analysis and multivariate logistic regression were used to screen the clinical parameters and imaging features of the training set,and the clinical imaging model was established.Finally,the clinical imaging model was combined with the radiomics model to establish a combined model.Results:(1)In the model for predicting the pathological grade of meningioma,there are 3 radiomics features used to build the radiomics model,and the classifier used is logistic regression.The AUC values in the training set and validation set are 0.81,0.79,respectively.Tumor shape is statistically significant in univariate and multivariate analysis(P < 0.05),and it is an independent predictor of meningeal pathological grade.When the tumor shape is used to identify the pathological grade of meningiomas,the AUC are 0.70 and0.60 in the training set and the validation set,respectively.In the combined model,the AUC values of the training set and the validation set are 0.87 and0.80,respectively.(2)In the model for predicting the Ki-67 expression level of meningioma,a total of 11 radiomics features are used to construct the radiomics model.The classifier used is support vector machine,and the AUC values of the training set and the validation set are 0.88,0.79,respectively.Clinical parameters and conventional imaging characteristics are not statistically significant in univariate analysis(P>0.05).Conclusion: Radiomics models based on CE-T1 WI performed well in preoperative prediction of the pathological grade and the Ki-67 expression level of meningioma,which has potential clinical application value.(2)In predicting the pathological grade of meningioma,the predictive performance of radiomics model and combined model is better than that of conventional imaging feature model.