BRAIN TUMOR DETECTION USING DEEP LEARNING

Abstract

The proposed project aims to develop an intelligent Brain Tumor Detection System using advanced deep learning techniques to address critical challenges in medical image analysis. The system analyzes MRI brain scan images and classifies them into four categories: Glioma, Meningioma, No Tumor, and Pituitary Tumor. By automating image classification, the project seeks to support healthcare professionals in making faster and more consistent assessments. The core component of the system is a custom Convolutional Neural Network (CNN) architecture, BrainTumorCNN_v2. The model consists of four convolutional blocks with filters ranging from 32 to 256, along with Batch Normalization and Dropout layers to improve learning stability and reduce overfitting. Global Average Pooling is incorporated to efficiently extract high-level image features. The model is trained using 150×150 pixel MRI images representing the four target classes. To improve model generalization and robustness, the system incorporates a comprehensive data augmentation pipeline. Random flipping, rotation, zooming, translation, and contrast adjustment are applied to training images, helping the model learn meaningful features despite variations in MRI image appearance and limited medical datasets. A modular Python-based prediction interface is also included. It accepts MRI images as input, loads the trained CNN model, processes the image, and generates predicted class probabilities. These outputs can provide valuable decision-support information for radiologists, clinicians, and healthcare stakeholders. Overall, the proposed system demonstrates the potential of deep learning and medical image analysis to assist brain tumor classification, improve diagnostic efficiency, and support data-driven healthcare. However, it is intended as a clinical decision-support tool rather than a replacement for professional medical diagnosis.