T2 Foundation Model

A Vision Transformer-based Masked Autoencoder Variational Autoencoder (MAE-VAE) for medical image analysis, specifically designed for T2-weighted MRI processing.

Features

• Vision Transformer (ViT) based architecture
• Masked autoencoding with variational inference
• NIfTI file support for medical imaging
• Slice-wise and volume-wise processing
• Advanced visualization tools

Installation

git clone https://github.com/Computational-Imaging-LAB/T2_FoundationModel.git
cd T2_FoundationModel
pip install -r requirements.txt

Usage

Training

python train.py --train_dir /path/to/train/nifti --val_dir /path/to/val/nifti

Inference

python inference.py --nifti_path /path/to/nifti/file.nii.gz

Documentation

Documentation is available at docs/build/html/index.html. To build the documentation:

cd docs
make html

License

MIT License

Citation

If you use this code in your research, please cite:

@software{t2_foundation_model,
  title = {T2 Foundation Model},
  author = {Computational Imaging Lab},
  year = {2025},
  url = {https://github.com/Computational-Imaging-LAB/T2_FoundationModel}
}