NeurIPS 2024: Bidirectional Recurrence for Cardiac Motion Tracking with Gaussian Process Latent Coding

This repository is the official implementation of "Bidirectional Recurrence for Cardiac Motion Tracking with Gaussian Process Latent Coding".

Requirements

Please use command

conda env create -f environment.yaml

to install the environment.

Data

• We used 3D Echocardiogram video for training: CardiacUDA dataset
• We used 4D cardiac MR images for training: ACDC dataset

Training

For CardiacUDA, to run training process:

python train_cardiauda.py

For ACDC, to run training process:

python train_acdc.py

Test

To run testing process:

python3 test_cardiacuda.py

or

python3 test_acdc.py

You need to replace the pretrained weight according to different datasets in files.

Acknowledgement

We would like to thank the great work of the following open-source project:

DiffuseMorph. FSDiffReg. DeepTag.

Citation

@article{yang2024bidirectional,
  title={Bidirectional Recurrence for Cardiac Motion Tracking with Gaussian Process Latent Coding},
  author={Yang, Jiewen and Lin, Yiqun and Pu, Bin and Li, Xiaomeng},
  journal={Advances in Neural Information Processing Systems},
  volume={37},
  pages={34800--34823},
  year={2024}
}