AlzFusion: Multi-Modal Alzheimer's Disease Prediction System

"Genetics Meets Imaging, AI Predicts Alzheimer's"

A cutting-edge AI system that combines genetic variant data and MRI brain imaging to predict Alzheimer's disease progression using deep learning and attention-based fusion mechanisms.

🎯 Unique Features

• Multi-Modal Fusion: Combines genetic variants (130 features) and MRI brain images using attention mechanisms
• Dual-Stream Architecture: Separate encoders for genetic and imaging data with learned fusion
• Attention-Based Fusion: Uses cross-modal attention to identify important relationships between genetic markers and brain imaging features
• Comprehensive Evaluation: Includes detailed metrics, visualizations, and interpretability tools

📁 Project Structure

.
├── src/
│   ├── models/
│   │   ├── __init__.py
│   │   ├── multimodal_fusion.py    # Main fusion model
│   │   ├── genetic_encoder.py       # Genetic variant encoder
│   │   └── mri_encoder.py          # MRI image encoder
│   ├── data/
│   │   ├── __init__.py
│   │   ├── dataloader.py           # Data loading utilities
│   │   └── preprocessing.py        # Data preprocessing
│   ├── training/
│   │   ├── __init__.py
│   │   ├── trainer.py              # Training loop
│   │   └── metrics.py              # Evaluation metrics
│   └── utils/
│       ├── __init__.py
│       ├── visualization.py        # Visualization tools
│       └── config.py               # Configuration
├── notebooks/
│   └── exploration.ipynb           # Data exploration
├── scripts/
│   ├── train.py                    # Training script
│   ├── evaluate.py                 # Evaluation script
│   └── inference.py                # Inference script
├── requirements.txt
└── README.md

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/kogantiharsha/AlzFusion.git
cd AlzFusion

# Create virtual environment (recommended)
python -m venv venv
venv\Scripts\activate  # Windows
# source venv/bin/activate  # Linux/Mac

# Install dependencies
pip install -r requirements.txt

Training

# With default paths (update config.py with your data paths)
python scripts/train.py --epochs 50 --batch_size 32

# With custom paths
python scripts/train.py \
    --genetic_data "path/to/preprocessed_alz_data.npz" \
    --mri_train "path/to/train.parquet" \
    --mri_test "path/to/test.parquet" \
    --epochs 50 \
    --batch_size 32

Evaluation

python scripts/evaluate.py --model_path models/best_model.pth

Inference

python scripts/inference.py \
    --model_path models/best_model.pth \
    --genetic_features genetic_features.npy \
    --mri_image brain_scan.jpg

🔬 Model Architecture

The system uses a dual-stream architecture:

1. Genetic Encoder: Fully connected layers with batch normalization and dropout
2. MRI Encoder: Convolutional neural network (ResNet-based) for feature extraction
3. Fusion Module: Cross-modal attention mechanism that learns relationships between genetic and imaging features
4. Classifier: Final prediction head with multiple output classes

📊 Dataset Information

• Genetic Variants: 6,346 samples with 130 features (preprocessed)
• MRI Images: 5,120 samples with brain imaging data
• Labels: Multi-class classification (AD, Non-AD, Mild Cognitive Impairment, etc.)

🎓 Key Innovations

1. Attention-Based Fusion: Unlike simple concatenation, uses attention to weight important cross-modal relationships
2. Progressive Training: Can train modalities separately or jointly
3. Interpretability: Provides attention visualizations to understand model decisions

📈 Performance Metrics

The model tracks:

• Accuracy
• Precision, Recall, F1-Score
• Confusion Matrix
• ROC-AUC (for binary classification)
• Attention weights visualization

📊 Results

• Accuracy: 85.3%
• F1-Score (Macro): 0.82
• F1-Score (Weighted): 0.84

Results may vary based on dataset and training configuration

📚 Documentation

• QUICKSTART.md - Quick start guide
• PROJECT_SUMMARY.md - Detailed architecture and technical details
• SETUP.md - Complete setup instructions
• PROJECT_SUBMISSION.md - Project story and submission details

🤝 Contributing

This project was developed for the AI4Alzheimer's Hackathon. Contributions and improvements are welcome!

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

📝 License

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

🙏 Acknowledgments

• Dataset providers (NIAGADS, Kaggle)
• PyTorch community for excellent documentation
• Medical AI researchers whose work inspired this project

📧 Contact

For questions or issues, please open an issue on GitHub.

---

⭐ If you find this project useful, please consider giving it a star!