[elastix] Anisotropic Rigidity

Stefan Klein s.klein at erasmusmc.nl
Tue Jun 21 17:56:01 CEST 2011

Hi Kannan,

> 1) I use tif files as input image. So I don't have voxel spacing info
> in that file. So I use a final grid spacing of 5x5x2 since the spacing
> of TIF file is 20 micron x 20 mixron x 50 micron. Correct me if I am
> wrong, I assumed this takes care of the image spacing issue.

I strongly recommend that you set the voxel spacing in the input images. 
Tif images are able to store the pixel spacing.

And then you should specify the bspline grid spacing in physical units 
(i.e. mm).

What you're doing now does not take care of the image spacing issue 
completely. It might be possible to get a similar result with some 
"hacks", but it's unclear what's going on, and the resulting deformation 
will be hard to interpret. It will save you a lot of headaches if you 
spend some time on getting the input image format correct. :)

> 2a) I used a input image where value of all voxels of slice "n" had
> values n just to visualize the contributing z-slices on the final
> images and saw that some where pulled from as far as +/-7 slices which
> doesn't happen anatomically (I apply the bspline trasnformation only
> and not the affine transformation). We would like some elasticity in
> z-dimension as well. Is it possible to use something as follows?
> (MovingImageDerivativeScales 1 1 0.2)

It is possible, but I don't think it will help much. It will just slow 
down the convergence.

> 2b) I currently use 2 transform parameter file, one for affine and one
> for bspline, let know if adding MovingImageDerivativeScales to the
> bspline param file affects the affine transformation too?

Adding MovingImageDerivativeScales to the bspline parameter files does 
not affect the affine registration.

> 3&4) SInce I use anisoptropic final grid spacing already, I hope my
> pyramid schedule will be anisotropic too, it this assumption correct?

There are two pyramids: one on the image data, and one on the bspline 
grid. If you set the final bspline grid spacing anisotropically, the 
lower resolution grids will also be anisotropic (unless specified 
otherwise using GridSpacingSchedule). But the ImagePyramidSchedule is by 
default isotropic: 8 8 8 4 4 4 2 2 2 1 1 1. So, at the first resolution, 
the image will be downsampled by a factor 8 in each dimension.

Kind regards,

More information about the Elastix mailing list