Introduction: Accurate grading of meningiomas is crucial for determining appropriate treatment strategies and predicting patient outcomes. Radiomics-based approaches have the potential to enable non-invasive and pre-operative grading of meningiomas. This study aims to develop and evaluate radiomics-based machine learning (ML) models for predicting meningioma grades using multiparametric MRI.
Methods: This study utilized the BraTS-MEN dataset, focusing on the released training set with available segmentations and grading information. A total of 701 patients (526 with grade 1 and 175 with grade 2-3 meningiomas) were included. The task was approached as a binary classification problem, distinguishing between low-grade (grade 1) and high-grade (grade 2-3) meningiomas. A total of 4872 radiomic features were extracted from each patient's T1, T1 with contrast, T2, and FLAIR MRI sequences using the PyRadiomics package. LASSO regression was employed to reduce the number of features to 170, which were then sorted based on feature importance. The data was split into training (60%), validation (20%), and test (20%) sets. TabPFN, XGBoost, LightGBM, CatBoost, and Random Forest algorithms were utilized to build predictive models. Hyperparameter tuning was done using Optuna and involved determining optimal model-specific parameters and number of features to be used. The validation set was used for tuning and calibration, while the test set was used for evaluation.
Results: The ML models demonstrated promising results in predicting meningioma grades using radiomic features. The area under the receiver operating characteristic curve (AUROC) was used as the primary metric. The CatBoost model achieved the highest mean AUROC of 0.782, indicating its ability to effectively distinguish between low-grade and high-grade meningiomas.
Conclusion : The radiomics-based ML approach presented in this study showcases the potential for non-invasive and pre-operative grading of meningiomas using multiparametric MRI. Further validation on larger and independent datasets is necessary to establish the robustness and generalizability of these findings.
.jpg)
