Skip to content
/ pTVreg Public

Deformable image registration (alignment) toolbox

License

Notifications You must be signed in to change notification settings

visva89/pTVreg

Repository files navigation

pTVreg is the deformable image registration (alignment) toolbox. It provides the following functionality:

Implemented image dissimilarity metrics:

  • local correlation coefficient (LCC)
    $D_{LCC}=1 - \frac{g_\sigma * (I\cdot J) - (g_\sigma*I)\cdot(g_\sigma*J)}{\sqrt{g_\sigma*I^2 - (g_\sigma*I)^2+\varepsilon }\,\sqrt{g_\sigma*J^2 - (g_\sigma*J)^2+\varepsilon }}$
  • nuclear (PCA) groupwise metric
    $D_{nuclear} =\|\; [\, I_1\circ T(k_1),\quad ..., \quad I_N\circ T(k_N)\, ]\; \|_\ast$
  • sum of squared differences (SSD)
  • sum of absolute differences (SAD)
  • normalized gradient fields (NGF)

Implemented regularizations:

  • temporal regularization
  • transformation Jacobian regularization
  • sparse and nonconvex spatial regularizations based on first- and second- order finite differences

The method works on CT, MRI, US and natural images. It is reported to be successful for registering breathing motion.

The method performs multiresolution (pyramidal) registration approach. On each resolution level it solves unconstrained optimization problem of the following form:

$\min_{k} D(J,\; I\circ T(k) ) + \lambda \text{TV}(k)$,

using LBFGS.

A Few Registration Examples

Figure: input images; registered images; tracking animation;

Figure: input images; registered images; tracking animation;

Figure: input images; registered images; tracking animation;

Figure: overlayed initial images; overlayed registered images;

Figure: registered images; tracking animation;

CT Lung Registration

We achieve state-of-the art registration result in DIR and COPD challenges.

4DCT 1 4DCT 2 4DCT 3 4DCT 4 4DCT 5 4DCT 6 4DCT 7 4DCT 8 4DCT 9 4DCT 10 Mean TRE Mean Time (sec.)
0.77 0.75 0.93 1.26 1.07 0.83 0.80 1.01 0.91 0.84 0.92 130
COPD 1 COPD 2 COPD 3 COPD 4 COPD 5 COPD 6 COPD 7 COPD 8 COPD 9 COPD 10 Mean TRE Mean Time (sec.)
0.71 1.91 0.77 0.67 0.71 0.66 0.75 0.78 0.64 0.85 0.8461 442

For details on lung registration please see examples and DIR and COPDgene pages.

pTV

If you decide to use our registration toolbox please consider referring to this repository or the following paper:

@article{vishnevskiy2017isotropic,
  title={Isotropic total variation regularization of displacements in parametric image registration},
  author={Vishnevskiy, Valery and Gass, Tobias and Szekely, Gabor and Tanner, Christine and Goksel, Orcun},
  journal={IEEE transactions on medical imaging},
  volume={36},
  number={2},
  pages={385--395},
  year={2017},
  publisher={IEEE}
}

The code is developed by Dr. Valery Vishnevskiy [email protected], Cardiac Magnetic Resonance group, Institute for Biomedical Engineering, ETH Zurich, University of Zurich, Computer-assisted Applications in Medicine, Computer Vision Laboratory, ETH Zurich.

ETHZurich ETHZurich

About

Deformable image registration (alignment) toolbox

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published