🧠 Brain Tumor Diagnosis Using Deep Learning

🔍 Overview

This project uses deep learning to accurately diagnose brain tumors from MRI scans. Our objective was to create a scalable, automated diagnostic tool to assist medical professionals in detecting various tumor types with high precision.

🧭 Approach

We worked with a dataset of 7,022 MRI images across four categories: glioma, meningioma, pituitary tumor, and no tumor. Multiple CNN-based architectures were trained to classify and localize tumors using advanced image processing techniques.

⚙️ Methodologies

• CNN: For classification and feature extraction
• U-Net: For medical image segmentation
• ResNet50: Deep network using residual connections
• VGG19: Pre-trained model for transfer learning
• Evaluation: Accuracy, precision, recall, F1-score, confusion matrix

🧰 Technologies

• Language: Python
• Libraries: TensorFlow, Keras
• Models: CNN, U-Net, ResNet50, VGG19
• Dataset: Br35H Brain MRI Dataset

💡 Key Learnings

• Learned model selection based on task-specific needs
• Improved preprocessing techniques for medical imaging
• Balanced precision and recall in sensitive diagnosis cases

📈 Results

VGG19 achieved an accuracy of 99.65% with a validation accuracy of 97.86%, followed by ResNet50 at 97.47%. Confusion matrices and evaluation metrics confirmed the model's effectiveness across all tumor classes, making it a reliable diagnostic tool in clinical settings.