Brain Tumor Classification using Deep Residual Networks

Abstract

Brain tumors represent one of the most critical and life-threatening forms of cancer worldwide, and accurate automated classification of MRI scans plays a crucial role in supporting timely diagnosis and treatment planning. In this study, a deep learning-based approach for automatic brain tumor classification is proposed, utilizing transfer learning with the ResNet50 convolutional neural network on a publicly available Kaggle dataset consisting of approximately 7,000 2D brain MRI images categorized into four classes: glioma, meningioma, pituitary tumor, and no tumor. The proposed approach employs a systematic two-stage fine-tuning strategy, in which all convolutional layers are initially frozen to preserve ImageNet feature representations, followed by selective unfreezing of the final convolutional block with a reduced learning rate to enable MRI-specific adaptation. Domain-justified data augmentation, including random horizontal flipping, rotation, and color jitter, is applied to improve robustness against real-world MRI variability. The model achieved an overall classification accuracy of 98.25% with a macro-averaged F1-score of 0.98 across all four tumor categories, demonstrating strong generalization and reliability. Grad-CAM visualization is additionally integrated to provide interpretability, validating that the model’s attention aligns with clinically relevant tumor regions. These results highlight the potential of interpretable deep transfer learning as a reproducible and accurate framework for AI-assisted brain tumor classification.