Optimizing Brain Tumor Classification through Innovative Texture-Based MRI Analysis: A Fusion of Novel GLCM and LBP with SVM Ensemble for Enhanced Accuracy

The exact categorization of brain tumors poses substantial difficulties in medical imaging due to complex tumor variations and diagnostic delays. This study presents a new method to improve texture-based information obtained from MRI scans for diagnosing brain malignancies, including gliomas, meningiomas, and pituitary tumors. Our approach involves combining distinctive features derived from the Gray-Level Co-occurrence Matrix (GLCM) with conventional Local Binary Pattern (LBP) histograms. This combination significantly enhances classification accuracy. Additionally, we generate new features through comprehensive feature engineering techniques, including creating aggregated features, interaction features, polynomial features, normalized features, and subtractive features. To address class imbalance, we employ the Random Over-sampling strategy, duplicating data points in the minority class to ensure equal representation. This strategy achieves a notable classification accuracy of 92.81%, surpassing the most cutting-edge existing methods and showcasing its potential to enhance the investigation of brain tumors in clinical environments.