We present two novel methods for domain-agnostic stroke lesion segmentation:

1. qATLAS: A neural network that estimates qMRI maps from standard MPRAGE images
2. qSynth: A direct synthesis approach for qMRI maps using label-conditioned Gaussian mixture models

Both methods leverage physics-based forward models to ensure physical plausibility in the simulated images.

To set up the required environment, run:

pip install -r requirements.txt

After installing dependencies, you can explore the available scripts:

python src/train_seg.py --help
python src/train_mprage.py --help

For detailed results and qualitative examples, please refer to our MICCAI 2025 paper.

We plan to release pretrained model weights in the near future. Stay tuned for updates.

If you use this code, please cite our work:

@misc{chalcroft2024domainagnosticstrokelesionsegmentation,
      title={Domain-Agnostic Stroke Lesion Segmentation Using Physics-Constrained Synthetic Data}, 
      author={Liam Chalcroft and Jenny Crinion and Cathy J. Price and John Ashburner},
      year={2024},
      eprint={2412.03318},
      archivePrefix={arXiv},
      primaryClass={eess.IV},
      url={https://arxiv.org/abs/2412.03318}, 
}