Multi-Level Brain Tumor Classification From MRI: A Review of Deep Learning Methods, Datasets, Evaluation, and Clinical Challenges

Abstract: Brain tumor classification from Magnetic Resonance Imaging (MRI) images has become an important area of research with the development of machine learning and deep learning techniques. This review examines the development of MRI-based brain tumor classification from early computer-aided and machine-learning approaches to recent deep- learning methods, with emphasis on Convolutional Neural Networks (CNNs), transfer learning, EfficientNet architectures, Vision Transformers, and hybrid CNN–Transformer models. The review analyzes representative studies, datasets, tumor-classification settings, evaluation measures, and reported findings across the 2000–2026 research period. Common classification categories include glioma, meningioma, pituitary tumors, benign/malignant classes, and no-tumor cases, while some studies also investigate tumor grades. The literature shows a methodological progression from handcrafted texture, shape, and intensity features with conventional classifiers toward end-to-end deep learning and transformer-based systems. However, differences in dataset size, class distribution, preprocessing, and validation strategies make direct comparison of reported performance difficult. Important challenges include limited dataset diversity, class imbalance, external validation, explainability, reproducibility, and clinical generalization. Future directions include larger multi-center datasets, independent validation, explainable artificial intelligence, lightweight models, privacy-preserving learning, multimodal analysis, and human-AI collaboration.

Keywords: Brain Tumor Classification; MRI; Deep Learning; CNN; Transfer Learning; Vision Transformer