Brain Tumor MRI Classification using Deep Learning

🧠 Project Overview

An advanced deep learning model is designed for binary classification of brain MRI scans to detect tumors (benign vs. malignant). This model leverages a custom attention-enhanced ResNet50 architecture, incorporating state-of-the-art training techniques to achieve high accuracy and reliability.

The development process involved benchmarking performance across several architectures, including ResNet18, ResNet50, EfficientNet, and Vision Transformer (ViT), to identify the optimal backbone for tumor detection. Furthermore, attention mechanisms like Attention Gate and Self-Attention were integrated to enhance feature refinement and improve the focus on critical tumor regions. A novel implementation of the Attention-Based Multiple Instance Learning (ABMIL) framework further boosts the model's interpretability and performance.

The model was trained and evaluated on the IAAA Challenge Dataset, demonstrating its robustness and adaptability to real-world medical imaging scenarios.

🌟 Key Features

• Custom attention mechanism for focused feature extraction
• ResNet50 backbone with attention gates
• Focal Loss for handling class imbalance
• Mixed precision training for improved performance
• Advanced data augmentation pipeline
• Comprehensive evaluation metrics and visualization

📊 Model Architecture

The model combines ResNet50 with custom attention gates:

• Base: Pretrained ResNet50
• Custom Components:
  • 3 Attention Gates (2048->1024, 2048->512, 2048->256)
  • Global Average Pooling
  • Custom classifier head (2048->512->1)
• Training Enhancements:
  • Focal Loss (alpha=0.25, gamma=2)
  • OneCycleLR scheduling
  • Mixed precision training with GradScaler

🛠 Requirements

torch>=1.7.0
torchvision>=0.8.0
albumentations>=0.5.2
opencv-python>=4.5.0
numpy>=1.19.2
pandas>=1.2.0
scikit-learn>=0.24.0
tqdm>=4.50.0
matplotlib>=3.3.0
imbalanced-learn>=0.8.0

🚀 Getting Started

Installation

# Clone the repository
git clone https://github.com/awrsha/Classification-of-brain-diseases.git
cd Classification-of-brain-diseases

# Install dependencies
pip install -r requirements.txt

Training

python 0.7914.py --batch_size 32 --epochs 60 --lr 1e-3

📈 Performance Metrics

The model is evaluated using multiple metrics:

• ROC-AUC
• Precision-Recall curves
• F1 Score
• Confusion Matrix
• Threshold Analysis

🔧 Model Configuration

Key hyperparameters:

{
    "batch_size": 32,
    "learning_rate": 1e-3,
    "weight_decay": 1e-2,
    "epochs": 60,
    "focal_loss_alpha": 0.25,
    "focal_loss_gamma": 2
}

🎯 Data Augmentation

Comprehensive augmentation pipeline including:

• Random rotation
• Flips
• Brightness/Contrast adjustments
• Coarse dropout
• Normalization
• Resize to 224x224

📊 Results Visualization

The project includes:

• ROC curves
• Precision-Recall curves
• Random sample predictions
• Confusion matrices
• Threshold analysis plots

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• The dataset providers
• The medical professionals who helped with data annotation
• The open-source community for the tools and libraries used

📧 Contact

For questions or feedback, please open an issue or contact [official.parvizi@gmail.com]