Segmented Glioma Classification Using Radiomics-Based Machine Learning: A Comparative Analysis of Feature Selection Techniques

Accurate classification of glioma grades is crucial for effective treatment planning and patient prognosis. In this extended study, we propose a comprehensive approach combining radiomics features and machine learning techniques to classify glioma grades. We explore the effectiveness of different feature selection methods, including Recursive Feature Elimination (RFE), Minimum Redundancy - Maximum Relevance (MRMR), and k-best, in identifying relevant features from segmented glioma for accurate classification. To achieve this, a deep learning approach that combines Convolutional Neural Networks (CNN) based on the hybridization of U-Net and SegNet is investigated in this study. The evaluation of the proposed approach involves training and testing machine learning models, including Linear Regression, Random Forest and XGBoost, using the selected features from each feature selection technique. The obtained results show that XGBoost with k-best feature selection achieves the highest accuracy and Area Under the Curve (AUC) for distinguishing between Low-Grade Gliomas (LGG) and High-Grade Gliomas (HGG). This indicates the effectiveness of the k-best feature selection method in capturing the most discriminative information for glioma grade classification. To the best of our knowledge, this is the first study to incorporate machine learning with RFE, MRMR, and k-best feature selection methods for predicting glioma grade. The proposed approach demonstrates improved accuracy compared to existing methods, highlighting the potential of radiomics and machine learning in glioma classification.