Hi Ed,
You can check the scaling and skew for a given transformation using
avscale. I did this for your matrices and, as desired, the first
introduces
an average scaling of 1.25 and minimal skew, while the second
(coplanar to hires) has 1.0 scaling and 0.0 skew - that is, a 6 DOF
transformation. Oh, and a 6 DOF transformation will in general not
have any zeros anywhere, so you can't use that.
As for the main problem of why the registration is wrong, I have one
suspicion which is that your voxel sizes might be incorrectly set in
one or more of your images. This is often the cause of images looking
the wrong size after registration - esp 6 dof ones. It is also hard to
tell
if the voxel size is wrong with certain viewing software, but such a
problem
can completely ruin a registration. To check what the voxel sizes are set
to have a look at the values of pixdim1 to pixdim3 with avwhd.
Also, uses slices to display the image and see if all views look sensible
and undistorted.
If there is a problem with your voxel size, you can fix your header with
avwcreatehd, and then the registration will hopefully work. If this is not
the problem, or the registration still doesn't work for some reason, then
please arrange for some way for me to access the data (don't send it
to the list). Your command lines look fine though, so I think it is most
likely to be a voxel size problem.
Cheers,
Mark
On Wednesday, November 13, 2002, at 10:02 AM, Edward Vessel wrote:
> Hi there -
>
> I have been using flirt to register a (skull stripped) hires image to the
> avg152T1_brain reference, and (skull stripped) low res coplanar epi
> image to
> the hires, and then combining the transforms to map the coplanar and other
> statisitical images into the standard space.
>
> I am finding that sometimes the transformation is very bad. In
> particular,
> the hires -> ref seems to work fine for one subject, but the coplanar
> registration is really bad. What I don't understand is that I specified
> that
> the coplanar to hires registration use only 6 degrees of freedom, so I'm
> not
> sure how the final image is getting so distorted. The final coplanar
> registration appears much larger, significantly offset, and stretched in
> one
> direction. The hires image seems fine.
>
> How can I tell if the registration is actually doing a rigid body
> transform
> or not?
>
> here are the commands I'm using and the .mat files generated:
>
> flirt -in s04_hiresanat_4FSL_brain.hdr -ref
> /usr/local/fmri_tools/fsl/etc/standard/avg152T1_brain.hdr -omat
> s04_hiresanat_norm.mat -bins 256 -cost corratio -searchrx -90 90 -searchry
> -90 90 -searchrz -90 90 -dof 12
>
> s04_hiresanat_norm.mat:
> 1.37998 -0.0448958 0.0541051 -13.686
> 0.0423571 1.10453 0.124788 -40.0922
> 0.00466688 -0.107659 1.26615 -107.039
> 0 0 0 1
>
> flirt -in s04_coplan_4FSL_brain.hdr -ref s04_hiresanat_4FSL_brain.hdr
> -omat
> s04_coplan_hires.mat -bins 256 -cost corratio -searchrx -90 90 -searchry
> -90
> 90 -searchrz -90 90 -dof 6
>
> s04_coplan_hires.mat:
> 0.996421 0.0210558 0.0818592 -30.3379
> -0.0194559 0.999605 -0.0202932 19.9357
> -0.0822541 0.0186279 0.996437 98.78
> 0 0 0 1
>
> (I guess I would have expected half of these to be exactly equal to
> zero if it were using a 6 parameter model)
>
>
> convert_xfm -matonly -concat s04_hiresanat_norm.mat -omat s04_coplan_norm.
> mat
> s04_coplan_hires.mat
>
> s04_coplan_norm.mat:
> 1.37146 -0.0148136 0.167787 -51.1022
> 0.0104516 1.10731 0.105396 -7.03109
> -0.0974012 -0.0839325 1.26421 15.7435
> 0 0 0 1
>
>
> thanks!
> Ed
>
> --
> Ed Vessel
> U. of Southern California [log in to unmask]
> Dept. of Neuroscience
> HNB, 3641 Watt Way http://geon.usc.edu/~vessel
> Los Angeles, CA 90089-2520
> (213) 740-6102
|