TCGA-BRCA ML+Gen AI Project

Overview

This project builds a comprehensive ML + Generative AI workflow using the TCGA-BRCA (Breast Invasive Carcinoma) dataset. The goal is to:

1. Predict Survival: Use Histopathology images (WSIs/Patches) to predict patient outcomes.
2. Multimodal Integration: Refine predictions by integrating clinical tabular data.
3. Preventive Gen AI: Generate "early-stage" disease images from late-stage samples to visualize disease progression "in reverse".

Project Structure

tcga-brca-genai/
├── data/
│   ├── raw/          # Original data (not committed)
│   ├── processed/    # Preprocessed patches and tensors
├── src/
│   ├── data/         # Data loaders and preprocessing
│   ├── models/       # Deep Learning architectures
│   ├── train/        # Training and validation loops
│   └── utils/        # Helper functions
├── notebooks/        # Experiments and analysis
├── docs/             # Methodology and citations
└── requirements.txt

Setup

1. Clone the repository.
2. Install dependencies:

   pip install -r requirements.txt

3. (Optional) Install OpenSlide binaries if on Mac:

   brew install openslide

Data

Data is sourced from the GDC Portal (TCGA-BRCA).

• Images: Whole Slide Images (WSIs) formatted as .svs.
• Clinical: Tabular data with survival information. Note: This project uses ROI patches extracted from WSIs for computational efficiency.

Methodology

(To be updated as the project progresses)

• Phase 1: Image-based survival modeling (CNN/ViT).
• Phase 2: Clinical + Image Fusion.
• Phase 3: Generative adversarial/diffusion models for stage translation.

📄 License

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

🙏 Acknowledgments

• The Cancer Genome Atlas (TCGA) for providing the BRCA dataset
• PyTorch and MONAI communities for excellent tools
• Research community for foundational work in computational pathology

📧 Contact

For questions or collaborations, please open an issue or contact [kmoralfig@gmail.com].

---

Note: This project is for research purposes only and should not be used for clinical decision-making without proper validation and regulatory approval.