🧠 Brain Tumor MRI Classification

A deep learning project that uses transfer learning and custom CNN architectures to classify brain tumors from MRI scans into four categories: Glioma, Meningioma, Pituitary, and No Tumor.

🎯 Project Overview

This project implements two different deep learning approaches:

1. Transfer Learning with Xception: Leveraging a pre-trained model for enhanced accuracy
2. Custom CNN: A dedicated convolutional neural network built from scratch

Both models are deployed through a user-friendly Streamlit web interface that provides:

• Real-time tumor classification
• Confidence scores
• Saliency maps for model interpretability
• AI-generated explanations of the model's focus areas

📊 Model Architecture & Performance

Transfer Learning Model (Xception)

• Pre-trained on ImageNet dataset
• 36 convolutional layers
• 21 million parameters
• Features:
  • Max pooling
  • Dropout layers for regularization
  • Softmax activation for classification
• Performance metrics:
  • High accuracy on test set
  • Robust against overfitting

Custom CNN Model

• 4 convolutional layers
• 4.7 million parameters
• Architecture:
  • Multiple Conv2D layers with ReLU activation
  • MaxPooling2D layers
  • Dropout for regularization
  • Dense layers with L2 regularization
  • Softmax output layer

🛠️ Installation

# Clone the repository
git clone https://github.com/stonewerner/brain-tumor-ML.git
cd brain-tumor-ML

# Install required packages
pip install -r requirements.txt

📦 Dependencies

• tensorflow >= 2.0
• streamlit
• numpy
• pandas
• pillow
• plotly
• opencv-python
• google-generativeai
• python-dotenv

🚀 Usage

Running the Web App

streamlit run app.py

🖥️ Web Interface Features

1. Image Upload: Support for jpg, jpeg, and png formats
2. Model Selection: Choose between Transfer Learning and Custom CNN
3. Visualization:
   • Original MRI scan
   • Saliency map highlighting model focus areas
4. Results Display:
   • Predicted tumor type
   • Confidence scores
   • Interactive probability chart
   • AI-generated explanation of the classification

📈 Data Processing

The project includes robust data handling:

• Image preprocessing and augmentation
• Brightness adjustment for training data
• Proper train/validation/test splits
• Standardized image sizing

🔍 Model Interpretability

• Saliency Maps: Visual explanation of model decisions
• Region Focus: Highlights critical areas in MRI scans
• AI Explanations: Generated using Google's Gemini model
• Confidence Metrics: Probability distribution across classes

👥 Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

📧 Contact

Stone Werner - stonewerner.com

Project Link: https://github.com/stonewerner/brain-tumor-ML