Leveraging Fractal Patterns in Machine Intelligence to Classify Brain Tumor Imaging

A significant global health problem is Brain tumor (BT), with patient survival becoming increasingly challenging in advanced stages. Early diagnosis and accurate classification of BTs into distinct categories is critical for accomplishing timely preventive measures. This study proposes a novel approach combining fractal analysis and machine intelligence (MI) for classifying MRI images of brain tumor. These images will be classified into multiple categories, including meningioma (MG), pituitary (PT), non-tumor (NT) and glioma (GM) types. To perform this multi-class classification of the said images, this proposed methodology uses a hybrid (or stack) approach where it combines the Logistic Regression (LRG) with the Neural Network (NNT) algorithms. In order to find out the effectualness of this proposed approach, we conducted a comparative analysis against several established and pre published ML methods, like K-Nearest Neighbor (KNNH), AdaBoost (ADB) and Random Forest (RFS). We also used individual LRG and NNT models as well as models like Decision Tree (DTR), Support Vector Machine (SVMN). We used Python-based Orange (3.26.0) software for implementing the comparative ML techniques and our proposed method. The study employs a dataset of 400 BTMIs obtained from Kaggle, comprising 100 images each of GM, MG, PT, and NT types. Performance evaluation when conducted using standard metrics (F1, RC, CA, PR) shows this suggested hybrid method outperforms RFS, ADB, KNNH and several other machine learning based approaches suggesting its superior capability in accurately classifying brain tumor MRI images across the specified categories.