Data source and study population

A groundbreaking study utilizing the BraTS-MEN dataset has paved the way for a non-invasive method to predict meningioma grade using pre-operative MRI scans. The dataset, contributed by academic medical centers across the United States, contained 698 patients with grade 1, 2, or 3 meningiomas. The study focused on developing a binary classification model to distinguish between low-grade and high-grade tumors based on radiomic features extracted from the MRI data.

The data was split into training, validation, and test sets for model development and evaluation. Feature extraction involved extracting 4872 radiomic features from each patient’s MRI sequences using the PyRadiomics package. To reduce dimensionality and mitigate overfitting, LASSO regression was employed, resulting in 176 predictive features for model development.

Five supervised machine learning algorithms were used to construct predictive models, with SMOTE applied to address class imbalance. Optuna was used for hyperparameter optimization, and model performance was evaluated on the test set using various metrics such as AUROC, precision, recall, and F1 score. The study also included visualization techniques like ROC curves, PRCs, calibration curves, and confusion matrices for model interpretation.

Overall, this study showcases the potential of machine learning in predicting meningioma grade from MRI scans, offering a promising tool for pre-operative planning and patient management. The computational efficiency of the methodology, coupled with the comprehensive analysis of radiomic features, highlights the significance of this research in the field of medical imaging and tumor classification.