Construction and validation of a machine learning-based automated prediction model for meningioma histopathologi-cal grading using MRI-enhanced images

CHEN Peng; SUN Miao; SONG Tingting; WANG Tianzuo; ZHANG Jinling

doi:10.20258/j.cnki.1006-9011.2025.10.008

Abstract：Objective To construct a machine learning-based model for automatic prediction of meningioma histopathologi-cal grading using MRI-enhanced images and validate the performance of segmentation and classification models.Methods A total of 126 patients with pathologically confirmed meningioma were enrolled,including 43 cases of high-grade meningioma and 83 cases of low-grade meningioma.Patients were randomly divided into a training set(70%,88 cases)and a test set(30%,38 cases),with an additional validation set(17 cases)randomly selected from the training set.A deep learning algorithm was used to train an automatic segmentation model,and the optimal model was selected based on the Dice coefficient.Radiomics features were subsequently extracted from the training set data,followed by multi-step feature selection to identify the optimal feature combination.Four machine learning algorithms were employed to construct classification models:Model A(feature extraction based on automated segmentation)and Model M(feature extraction based on manual segmentation).Finally,the optimal classifi-cation models were evaluated and compared using metrics including ROC curves,calibration curves,decision curve analysis,and the net reclassification index(NRI).Results Model A demonstrated excellent performance across the training set,valida-tion set,and test set,with mean Dice coefficients of 0.915,0.886,and 0.871,respectively.Among the classification models,the random forest algorithm yielded the best performance.ROC analysis showed that Model A achieved an AUC of 0.851(95%CI:0.713-0.989),while Model M had an AUC of 0.831(95%CI:0.683-0.979),with no statistically significant difference be-tween the two(P=0.325).Calibration curves indicated good model fit for both models.Decision curve analysis revealed that Model A exhibited slightly higher standardized net benefit across most threshold probabilities.Sensitivity,specificity,accuracy,and AUC were consistent between the two models when predicting the same test set,NRI=0.Conclusion The proposed ma-chine learning model integrating deep learning and radiomics demonstrates robust performance in predicting histopathological grading of meningioma using MRI-enhanced images.This approach provides a valuable tool for preoperative assessment and treat-ment decision-making.