Slice-to-Volume Registration Transformer (SVoRT)

This repo is the official implementation of the paper 'SVoRT: Iterative Transformer for Slice-to-Volume Registration in Fetal Brain MRI'

Resources

• Paper ( Springer | arXiv )
• Check out NeSVoR for an application of SVoRT

Requirements

• python 3.9
• pytorch 1.10
• pyyaml
• scikit-image
• antpy
• scipy
• nibabel

Training from scratch

Generate training data

To generate training data, please download the CRL atlas and FeTA dataset v2.1, unzip them in dataset/, and run preprocessing.py. You may also add your own training data (see RegisteredDataset in .src/data/dataset.py).

Modify hyperparameters

The hyperparameters of data simulation and model are stored in ./src/config/.

Run the training script

python train.py --config ./config/config_SVoRTv2.yaml \
                --output ../results/SVoRTv2

Pretrained model

To use the pretrained model, please first download the pretrain weights.

Testing

python test.py --config ./config/config_SVoRTv2.yaml \
               --output ../results/SVoRTv2/test_output \
               --checkpoint ../results/SVoRTv2/checkpoint.pt

Citation

@inproceedings{xu2022svort,
  title={SVoRT: Iterative Transformer for Slice-to-Volume Registration in Fetal Brain MRI},
  author={Xu, Junshen and Moyer, Daniel and Grant, P Ellen and Golland, Polina and Iglesias, Juan Eugenio and Adalsteinsson, Elfar},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={3--13},
  year={2022},
  organization={Springer}
}

Contact

For questions, please send an email to [email protected]