A Cloud-Based Approach to Brain Tumor Classification Using Convolutional Neural Networks

Abstract

This work presents a cloud-based brain tumor classification system leveraging Convolutional Neural Networks (CNNs) to enhance diagnostic accuracy and transparency in medical imaging. The system classifies MRI images into four categories: Glioma, Meningioma, Pituitary, and No Tumor. Multiple established CNN architectures—including EfficientNet, MobileNet, and ResNet—were implemented using PyTorch to evaluate performance. The best results were achieved with EfficientNet-B1, which attained a test accuracy of 99.43% and a test loss of 0.0264 on a dataset of approximately 3,000 images.

To enhance trust and clinical usability, Explainable AI (XAI) techniques such as Grad-CAM were integrated, providing visual interpretability of the model’s predictions. The entire system was deployed using cloud-based technologies, including Google Colab for model training, Flask for web-based user interaction, and Microsoft Azure for model deployment and MongoDB for data transection. This cloud-native setup ensures real-time inference, scalability, and remote accessibility without the need for on-premise infrastructure.

Overall, this work demonstrates the potential of combining cloud computing, deep learning, and model interpretability to support reliable and efficient brain tumor diagnosis in modern healthcare environments.