fireants 1.5.0

FireANTs: Adaptive Riemannian Optimization for Multi-Scale Diffeomorphic Registration

:fire: FireANTs: Adaptive Riemannian Optimization for Multi-Scale Diffeomorphic Registration

The FireANTs library is a lightweight, general-purpose registration package for Riemannian diffeomorphic registration on GPUs. It is designed to be easy to use, fast, accurate, and extensible. The salient features of the library include:

• Powerful optimization algorithms ⚡️🧠: FireANTs utilizes a novel Riemmanian Adaptive optimization algorithm to perform gradient descent on the space of diffeomorphisms. Diffeomorphisms are a deformable transforms that do not lie on a Euclidean manifold, and hence require specialized optimization algorithms.
• Blazing fast inference 🚀: FireANTs is much faster at inference than most other registration libraries, including deep learning methods.
• Lightweight 🧩: FireANTs consumes about 10x less memory than state-of-the-art deep learning methods, scaling easily to high resolution images.
• Batched Registration 📚: Simultaneously register multiple image pairs with ease, which reduces runtime per image pair to upto 250ms per image pair on a single GPU.

• Composable transforms 🔗⚡️: FireANTs provides a simple API for composing transforms. You can easily compose multiple transforms and apply the composition to images directly. This avoids aliasing due to multiple resampling. Check out the documentation for more details.

FireANTs achieves up to 10x speedup compared to traditional registration methods and deep learning approaches while improving accuracy and runtime.

Installation

To use the FireANTs package, you can either

• Use Docker (see docker/README.md, highly recommended)
• Clone the repository and install the package locally
• Install the package directly from PyPI.

If you do not want to use Docker, we recommend using a fresh Anaconda/Miniconda environment to install the package.

conda create -n fireants python=3.9

To install FireANTs locally:

git clone https://github.com/rohitrango/fireants
cd fireants
pip install .
cd fused_ops && python setup.py build_ext && python setup.py install && cd ..

By default, doing pip install . will NOT include the fused CUDA operations. I highly recommend installing the fused CUDA operations using the command above.

Or install from PyPI (does not install fused ops):

pip install fireants

🤷 Tutorial

To check out some of the tutorials, check out the tutorials/ directory for usage. Alternatively, to reproduce the results in the paper checkout the fireants/scripts/ directory.

🚀 Fused CUDA Operations

If you want to use the fast and memory efficient fused CUDA operations, you can install the fireants_fused_ops package. See fused_ops/README.md for a basic user guide.

👾 CLI Tools

FireANTs provides command-line interface tools similar to the original ANTs toolkit. For detailed instructions and available tools, see cli/README.md.

📚 Template building

FireANTs includes a powerful template builder for creating anatomical templates from medical images. For detailed instructions, configuration options, and usage examples, see fireants/scripts/template/README.md.

📖 Documentation

You can also check out the Documentation. Feel free to reach out to me for improvements in the documentation. To build the documentation locally, run the following command from the root of the repository:

mkdocs serve -f docs/mkdocs.yml

📊 Datasets

In the paper, we use the datasets as following:

• Klein's evaluation of 14 non-linear registration algorithms: here
• EMPIRE10 lung registration challenge: here
• Expansion Microscopy dataset: here

🤝 Contributing

Feel free to add issues or pull requests to the repository. We welcome contributions to the package.

🧪 Tests

To run the tests, run the following command:

bash run_tests.sh

Feel free to report any issues with the tests to the issues page.

License

Please refer to the LICENSE file for the license details, especially pertaining to redistribution of code and derivative works.

Citation

If you use FireANTs in your research, please cite the following papers:

@article{jena2024fireants,
  title={FireANTs: Adaptive Riemannian Optimization for Multi-Scale Diffeomorphic Registration},
  author={Jena, Rohit and Chaudhari, Pratik and Gee, James C},
  journal={Nature Communications},
  year={2024}
}

@inproceedings{jena2025scalable,
  title={A Scalable Distributed Framework for Multimodal {GigaVoxel} Image Registration},
  author={Jena, Rohit and Zope, Vedant and Chaudhari, Pratik and Gee, James C},
  booktitle={The Fourteenth International Conference on Learning Representations},
  year={2026},
  url={https://openreview.net/forum?id=8dLexnao2h}
}

If you use FireANTs-as-a-layer (Deep Implicit Optimization, code), cite the following paper:

@article{jena2025deep,
  title={Deep implicit optimization enables robust learnable features for deformable image registration},
  author={Jena, Rohit and Chaudhari, Pratik and Gee, James C},
  journal={Medical Image Analysis},
  volume={103},
  pages={103577},
  year={2025},
  publisher={Elsevier}
}

The following evaluation studies validate FireANTs against other state-of-the-art methods:

@inproceedings{jena2026the,
  title={The LU-Mirage - An independent evaluation of the zero-shot claims in the {LUMIR} challenge},
  author={Jena, Rohit and Chaudhari, Pratik and Gee, James},
  booktitle={Medical Imaging with Deep Learning - Validation Papers},
  year={2026},
  url={https://openreview.net/forum?id=a46yLtToIK}
}

@inproceedings{jena2024deep,
  title={Deep Learning in Medical Image Registration: Magic or Mirage?},
  author={Jena, Rohit and Sethi, Deeksha and Chaudhari, Pratik and Gee, James C.},
  booktitle={Advances in Neural Information Processing Systems},
  volume={37},
  pages={108331--108353},
  year={2024},
  url={https://proceedings.neurips.cc/paper_files/paper/2024/file/c3fe2a07ec47b89c50e89706d2e23358-Paper-Conference.pdf}
}