Brain Tumor Classification Using Transfer Learning with EfficientNet

Abstract

This project presents a cloud-based system that uses deep learning to automatically classify brain tumors from MRI images. The system is built using a Convolutional Neural Network (CNN) model designed for multi-class tumor detection. To improve the accuracy and reliability of the model, techniques such as residual connections, batch normalization, and attention mechanisms are used for better feature extraction and generalization.The trained model is deployed using a FastAPI backend service and connected to a React-based frontend, allowing users to upload MRI images and receive real-time predictions. The system provides probability scores and confidence levels for each prediction, along with visual explanations using Grad-CAM heatmaps, which highlight the regions of the MRI image that influenced the model’s decision.The model was trained using MRI datasets collected from multiple institutions and achieved an accuracy of 97.9% with an F1-score of 0.978. The system can process each MRI scan in approximately 300 milliseconds, making it suitable for real-time use. In addition, the platform provides RESTful APIs so it can be integrated into hospital radiology systems, along with dashboards for monitoring and analyzing data over time.