BRAIN TUMOR DETECTION AND CLASSIFICATION FROM MRI IMAGES: DEEP LEARNING PERSPECTIVES AND AUTOMATED DIAGNOSTIC APPROACHES

Abstract

One of the most serious problems that exist today in the field of neurology is brain tumors, with early and correct diagnosis being essential for effective treatment and survival of patients. Manual analysis of Magnetic Resonance Imaging scans take much time, depends on the experience of an individual who analyses it, and requires special radiology skills, which becomes more challenging. This research work addresses these challenges through two interconnected contributions under the broader perspective of deep learning based automated brain tumor detection.A detailed and exhaustive literature review about automated brain tumor detection techniques developed during the period from 2020 to 2026 is presented in one contribution. It involves classical Machine Learning models, deep CNN architectures, transfer learning models, object detectors like YOLOv5v, Vision transformers and hybrid models. These frameworks are carefully studied and critically compared to their design, popular datasets such as BraTS, Figshare, Kaggle Brain MRI dataset and evaluation criteria. Systematically identified limitations include computational efficiency, class imbalance, explainability of the model, lack of annotated data and differences between clinical and benchmark data. Another contribution describes the development of the VGG16 transfer learning architecture for four-class brain tumor classification involving glioma, meningioma, pituitary and no tumor class using a publicly available database of 7,023 brain MRI images.The model presented in this paper uses data augmentation techniques and a custom classifier head to achieve 96% test accuracy with weighted average precision, recall, F1-score of 0.96. This research offers not only a resourceful summary but also a solid baseline approach towards AI-based brain tumor diagnosis.