We study the influence of (learned) landmark correspondences on intensity-based deformable image registration involving "hard" organs like the lung and liver. Our work extends the self-supervised model developed by Grewal et al. (2023) by proposing the use of a mask during training to focus the model on key anatomical structures (e.g. vessels inside the liver).
We demonstrate the benefits our soft mask extension using two use-cases:
- Lung CT registration (using the 4DCT and COPDgene datasets)
- Liver lesion co-localization (using a dataset of dynamic contrast-enhanched (DCE) MR images collected at UMC Utrecht)
Use the following to clone the repository and install packages.
git clone https://github.com/ishaanb92/LandmarkBasedRegistration.git
python setup.py install
You will also need to install Elastix yourself from here. Set the elastix_path and transformix_path to the paths where you installed the binaries for elastix and transformix when using the ElastixInterface and TransformixInterface classes during registration.
If you use this code in your work, you can cite the following publication:
Plain text:
Bhat I, Kuijf HJ, Viergever MA, Pluim JPW. Influence of learned landmark correspondences on lung CT registration. Med Phys. 2024; 1-16. https://doi.org/10.1002/mp.17120
BibTex:
@article{https://doi.org/10.1002/mp.17120,
author = {Bhat, Ishaan and Kuijf, Hugo J. and Viergever, Max A. and Pluim, Josien P. W.},
title = {Influence of learned landmark correspondences on lung CT registration},
journal = {Medical Physics},
volume = {n/a},
number = {n/a},
pages = {},
keywords = {deep learning, image registration, landmark correspondence},
doi = {https://doi.org/10.1002/mp.17120},
url = {https://aapm.onlinelibrary.wiley.com/doi/abs/10.1002/mp.17120},
eprint = {https://aapm.onlinelibrary.wiley.com/doi/pdf/10.1002/mp.17120},
}