visva89 / pTVreg

Deformable image registration (alignment) toolbox

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

• 2D and 3D image registration
• pairwise and groupwise registration
• cubic and linear spline motion parametrization
• multichannel data
• simple model for simultaneous registration and segmentation (via atlas deformation)
• approximate 2D registration run time: 5 sec (8 images), 40 sec (60 images) on a 6 core CPU
• approximate 3D registration run time: 90 sec pair of 226x226x106 volumes on a 6 core CPU
• Matlab implementation, compiled MEX files for Linux, Windows, and OSX (no multithreading)
• Please see the complete set of examples here
• Documentation is available here
• Troubleshooting guide
• For more details please refer to "Vishnevskiy V, Gass T, Szekely G, Tanner C, Goksel O. Isotropic total variation regularization of displacements in parametric image registration. IEEE TMI. 2017."

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 [e-mail], Cardiac Magnetic Resonance group, Institute for Biomedical Engineering, ETH Zurich, University of Zurich, Computer-assisted Applications in Medicine, Computer Vision Laboratory, ETH Zurich.