Skip to main content

Digital diffeomorphism: Non-diffeomorphic volume and non-diffeomorphic area computation

Project description

Digital Diffeomorphism: Non-diffeomorphic volume and Non-diffeomorphic area

This is an implementation of the non-diffeomorphic volume and non-diffeomorphic area computation we introduced in our paper:

Liu, Yihao, et al. "On Finite Difference Jacobian Computation in Deformable Image Registration." arXiv preprint arXiv:2212.06060 (2022).

Motivation

The Jacobian determinant $|J|$ of spatial transformations is a widely used measure in deformable image registration, but the details of its computation are often overlooked. Contrary to what one might expect, when approximated using the central difference, the number of pixels/voxels with negative $|J|$ does not reflect if the transformation is diffeomorphic or not.

The example above demonstrate a failure case of the central difference based $|J|$. The center pixel has central difference based $|J|=1$ but it is not diffeomorphic. In fact, the transformation at the center pixel has no effect on the computation of central difference based $|J|$, even if it moves outside the field of view.

We proposed the definition of digital diffeomorphism that solves several errors that inherent in the central difference based $|J|$. We further propose to use non-diffeomorphic volume (NDV) and non-diffeomorphic area (NDA) to measure the irregularity of 3D/2D transformations. The proposed NDV and NDA also measures the severity of the irregularity whereas the central difference approximated $|J|$ is only a binary indicator.

The center pixel in all three cases shown above would be considered diffeomorphic because of the checkerboard problem. The forward difference based $|J|$ is able to identiy that (b) and (c) exhibit folding, but only NDA can provide the observation that the non-diffeomorphic space caused by the transformation shown in (c) is larger than (b). The transformation shown in (b) is obviously more favorable than the one shown in (c) in terms of regularity. As such it is important for us to be able to draw distinctions between there two senarios.

Getting Started

Installation

The easiest way to install the package is through the following command:

pip install digital-diffeomorphism

To install from the source:

git clone https://github.com/yihao6/digital_diffeomorphism.git

cd digital_diffeomorphism

python setup.py install

Usage

To evaluate a 3D sampling grid with dimension $H\times W\times D\times 3$

ndv grid_3d.nii.gz

This will calculate

  1. non-diffeomorphic volume; and
  2. non-diffeomorphic voxels computed by the central difference.

If the transformation is stored as a displacement field:

ndv disp_3d.nii.gz --disp

To evaluate a 2D sampling grid with dimension $H\times W\times 2$

nda grid_2d.nii.gz

This will calculate

  1. non-diffeomorphic area; and
  2. non-diffeomorphic pixels computed by the central difference.

If the transformation is stored as a displacement field:

ndv disp_2d.nii.gz --disp

Example inputs can be found at https://iacl.ece.jhu.edu/index.php?title=Digital_diffeomorphism

Potential Pitfalls

  1. Several packages implement spatial transformations using a normalized sampling grid. For example, torch.nn.functional.grid_sample. In this package, we use un-normalized coordinates to represent transformations. Therefore, the input sampling grid or displacement field should be in voxel or pixel units. In case the input is normalized, it must be unnormalized prior to using this package.

Citation

If you use this code, please cite our paper.

@article{liu2022finite,
  title={On Finite Difference Jacobian Computation in Deformable Image Registration},
  author={Liu, Yihao and Chen, Junyu and Wei, Shuwen and Carass, Aaron and Prince, Jerry},
  journal={arXiv preprint arXiv:2212.06060},
  year={2022}
}

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

digital_diffeomorphism-1.0.1.tar.gz (11.1 kB view details)

Uploaded Source

Built Distribution

digital_diffeomorphism-1.0.1-py3-none-any.whl (10.4 kB view details)

Uploaded Python 3

File details

Details for the file digital_diffeomorphism-1.0.1.tar.gz.

File metadata

File hashes

Hashes for digital_diffeomorphism-1.0.1.tar.gz
Algorithm Hash digest
SHA256 1c962235326196959a222696984585dfe8ffe2fc621a087fb3a8ccb24edb5b02
MD5 6d445d0adeacbaf40dcd933fe80462d3
BLAKE2b-256 83e874d5eb66b5ad4746e0b816143ca13294183c873ecc14676f4b0b956f165c

See more details on using hashes here.

File details

Details for the file digital_diffeomorphism-1.0.1-py3-none-any.whl.

File metadata

File hashes

Hashes for digital_diffeomorphism-1.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 3034be0034ef939bb0296bdfe70f2003b8cdb44ad3ded833471c6363992b46be
MD5 62339cd035bb808e0a228351a4ac1ff2
BLAKE2b-256 86a021216fed37f9fbb93ea724ee8ccd59d2b0b00642b80637ba4e45366bdc10

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page