BRAIN TUMOR DETECTION AND CLASSIFICATION USING DEEP LEARNING

Abstract

Brain tumor detection from magnetic resonance imaging (MRI) scans is a critical and challenging task in medical image analysis, where early and accurate diagnosis plays a vital role in improving patient survival rates and treatment planning. This study proposes an automated brain tumor detection and classification framework based on deep learning techniques, incorporating a basic Convolutional Neural Network (CNN), VGG-16, and ResNet-101 architectures.

The proposed models are trained and evaluated on benchmark MRI brain tumor datasets using transfer learning and optimized training strategies to address challenges associated with limited data availability. Each model is trained for 20 epochs in a GPU-accelerated environment. Performance evaluation is conducted using standard classification metrics, including accuracy, loss, precision, recall, and F1-score. The experimental results demonstrate that all three models achieve strong classification performance. Among them, ResNet-101 achieves the highest performance, with a training accuracy of 99.94% and a testing accuracy of 98.31%. VGG-16 follows closely, achieving 99.21% training accuracy and 97.85% testing accuracy, while the basic CNN model obtains 98.74% training accuracy and 94.31% testing accuracy. The corresponding loss values remain consistently low across all models, indicating effective learning behavior and strong generalization capability. Furthermore, the training and validation curves illustrate stable convergence across epochs, with minimal evidence of overfitting. These results highlight the effectiveness of transfer learning in enhancing feature extraction from MRI scans and improving classification performance. To further improve model robustness, data augmentation techniques and pretrained architectures are employed, particularly to mitigate the limitations posed by small dataset sizes. Overall, the proposed system demonstrates significant potential for assisting medical professionals in the early detection of brain tumors by providing fast, accurate, and reliable predictions. The findings suggest that deep learning architectures, particularly ResNet-101 and VGG-16, serve as effective tools for automated brain tumor classification and can be extended for deployment in real-world clinical decision-support systems.