Hybrid machine and deep learning in brain tumor MRI analysis

Abstract

This study examines hybrid machine learning and deep learning methods for classifying brain tumors using Magnetic Resonance Imaging (MRI). We review recent advances that often combine convolutional neural networks (CNNs) with traditional classifiers, with accuracies ranging from 95% to over 99%. The review addresses brain tumor types, standardized datasets, and performance evaluation metrics. Evidence suggests that feature fusion strategies, which integrate deep learning representations with custom radiomic features, consistently achieve superior results. Still, computational complexity, limited dataset availability, and the need for clinical validation remain significant barriers to widespread implementation. In addition to the review, we provide a lightweight baseline experiment on a widely used public dataset (Kaggle Brain Tumor MRI) using handcrafted features with SVM classifiers. This baseline serves as a reproducible reference point that contextualizes the performance-complexity trade-offs discussed in the reviewed literature and highlights remaining gaps toward clinical deployment. The paper concludes by recommending future research to develop more efficient, interpretable, and clinically applicable brain tumor classification systems to support diagnostic decision-making in healthcare.