Home

Awesome

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

Requirements

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 junshen@mit.edu