Using MRI-Based machine Learning model for distinguishing Meningothelial and Transitional Meningiomas : A Retrospective Study

Objective Meningiomas constitute a considerable portion of central nervous system tumors, with the most common variants being meningothelial and transitional types. However, the long-term management approaches and prognostic outcomes after surgery for the two diseases exhibit notable differences.This study aimed to develop and validate MRI-based radiomics models for the preoperative differentiation of these two subtypes. Materials and methods A total of 64 patients were analyzed, comprising 34 patients with meningothelial and 30 patients with transitional meningiomas from 2018 to December 2022. Radiomics features were extracted from MRI images, and feature selection was performed through the Mann–Whitney U test and mRMR. The prediction performance of radiomics signatures was assessed by AUC value and decision curve analysis (DCA). Results The most effective model, combining Stepwise Generalized Linear Model (Stepglm) with backward selection and Elastic Net (Enet) (alpha = 0.4), achieved an accuracy of 0.906 and an AUC of 0.844, showing the good discriminative ability, and DCA showed favorable predictive performance of the MRI-based radiomics signatures. In addition, the clinical efficacy of this machine learning models markedly exceeds that of conventional imaging feature which yielded an AUC of 0.64. In contrast, the combined model that incorporates both traditional imaging features and radiomic features exhibits comparable efficacy compared to a machine learning model based solely on radiomics. Conclusion These findings suggest that MRI-based machine learning alone holds significant promise in differentiating between meningothelial and transitional meningiomas, highlighting its potential for non-invasive histological prediction and individualized neurosurgical planning.