SlicerUniGradICON

This is the official Slicer Extension for uniGradICON: A Foundation Model for Medical Image Registration, and multiGradICON: A Foundation Model for Multimodal Medical Image Registration. The extension provides a Slicer interface for the models, allowing users to perform image registration tasks using the models. The official repository for the models can be found here.

uniGradICON: A Foundation Model for Medical Image Registration
Tian, Lin and Greer, Hastings and Kwitt, Roland and Vialard, Francois-Xavier and Estepar, Raul San Jose and Bouix, Sylvain and Rushmore, Richard and Niethammer, Marc
MICCAI 2024 https://arxiv.org/abs/2403.05780

multiGradICON: A Foundation Model for Multimodal Medical Image Registration
Demir, Basar and Tian, Lin and Greer, Thomas Hastings and Kwitt, Roland and Vialard, Francois-Xavier and Estepar, Raul San Jose and Bouix, Sylvain and Rushmore, Richard Jarrett and Ebrahim, Ebrahim and Niethammer, Marc
MICCAI Workshop on Biomedical Image Registration (WBIR) 2024 https://arxiv.org/abs/2408.00221

Please (currently) cite as:

@article{tian2024unigradicon,
  title={uniGradICON: A Foundation Model for Medical Image Registration},
  author={Tian, Lin and Greer, Hastings and Kwitt, Roland and Vialard, Francois-Xavier and Estepar, Raul San Jose and Bouix, Sylvain and Rushmore, Richard and Niethammer, Marc},
  journal={arXiv preprint arXiv:2403.05780},
  year={2024}
}

@article{demir2024multigradicon,
  title={multiGradICON: A Foundation Model for Multimodal Medical Image Registration},
  author={Demir, Basar and Tian, Lin and Greer, Thomas Hastings and Kwitt, Roland and Vialard, Francois-Xavier and Estepar, Raul San Jose and Bouix, Sylvain and Rushmore, Richard Jarrett and Ebrahim, Ebrahim and Niethammer, Marc},
  journal={arXiv preprint arXiv:2408.00221},
  year={2024}
}

Installation

This extension depends on Pytorch Slicer extension. If it is not installed, please install it from Extension Manager by simply searching for "Pytorch".

To install the uniGradICON extension, clone this repository using the following command:

git clone https://github.com/uncbiag/SlicerUniGradICON.git

Then, drag and drop the cloned repository folder into the Slicer window and click "OK" to install the extension. After the installation is done, the extension will appear under the Modules -> Registration menu.

User Guide

1. Initially, select the fixed and moving images, and their modality (MR or CT/CBCT). The extension will preprocess the images based on the selected modality. The user does not need to perform any preprocessing steps manually.

2. The extension provides two models: uniGradICON and multiGradICON. Both models have the same interface and can be used to perform image registration tasks. The user can select the desired model from a drop-down menu.

3. Additionally, the user can specify the number of iterations for the instance optimization (IO steps) process. The default value is set to 0 iterations. However, we recommend using at least 50 iterations for more challenging registration tasks.

4. The user can also select the similarity loss function that will be used during the optimization process. The available similarity losses are:

   • Localized Normalized Cross-Correlation (LNCC)
   • Squared Localized Normalized Cross-Correlation (Squared LNCC)
   • Modality Independent Neighborhood Descriptor (MIND - SSC)

   For multimodal registration tasks, we recommend using the Squared LNCC or MIND - SSC similarity loss. Note that if IO steps are set to 0, the selected similarity loss function will not be applied.

5. The user can select the device that model will be run on. The available options are:

   • CPU
   • GPU (CUDA)

   If GPU is selected, the extension will use the GPU for the optimization process. Note that the GPU option is only available if the user has a CUDA-compatible GPU. Also, there are some limitations on the GPU memory usage. The peak memory usage of the model is as follows:

   • w/o Instance Optimization: 3364.0 MB
   • w/ Instance Optimization using LNCC: 11446.0 MB
   • w/ Instance Optimization using Squared LNCC: 11490.0 MB
   • w/ Instance Optimization using MIND-SSC: 14814.0 MB

   Please make sure that your GPU has enough memory to run the model. If the GPU memory is not sufficient, prefer using the CPU option.

6. Finally, the user can select the output transform and transformed image. These will be saved under the specified names in the Slicer scene.

7. After selecting all necessary parameters, the user can click the "Run Registration" button to initiate the registration process. The extension will perform the task using the selected model and parameters. Depending on the selected parameters and CUDA availability, the process may take a few minutes.

8. The user can visualize the registration results by selecting the output transform and transformed image under the "Data" module. Also, we recommend using the "CheckerBoard Filter" module to compare the registration result with the fixed image.

TroubleShooting

Here are some common issues that may arise during the installation and use of the extension, along with their solutions:

• On first use, the extension downloads the required PyTorch libraries. This installation may take a few minutes, and it is normal for the Slice interface to appear frozen during this process. Please wait until the installation is complete. You can monitor the progress in the Slicer Python Console.

• The I/O process takes approximately one second per iteration on a GPU. Please be patient during the optimization process, particularly when using a CPU.

• If an error arises due to incompatibility between setuptools and distutils, update setuptools using the following command in the Slicer Python Console: slicer.util.pip_install("setuptools -U")

Acknowledgements

This implementation is based on the SlicerANTs and SlicerSegmentWithSAM repositories. We thank the authors for their work.