> I am planning to invert the spatial normalisation (sn.mat-file) for the
> following aim:
> I want to bring my MRI in Transmission image (TX) space (scans of the same
> person), therefore both images will be first spatially normalised to their
> respective templates. Then I want to invert the sn.mat file obtained from
> the TX normalisation to apply it to the MRI.
First of all, make sure that the MRI and transmission image are in register.
Use the Coregister button, which should do within subject multi-modal
coregistration. Check the results with the Check Reg button. This
coregistration only works for data that has not been spatially normalised.
Assuming that the data are in register, then you can estimate a deformation
from one volume, and apply it to another. Therefore instead of seperately
registering the MRI to an MRI template, and the transmission image to the tx
template, you would instead (probably) match the MRI to an MRI template and
apply the same transformation to the transmission scan.
> Here my questions:
> 1. The resulting MRI in TX space is not exactly in register (actually, it
> is after subsequently applying a realignment; might it be possible that I
> have to invert the affine transformation accompanied in the sn-file as
> well?): I gave as basis for the inversion the TX-image before it has been
> normalised to the template: Is this right?
Correct. As a reminder, the procedure for creating un-normalised images is as
follows.
1) Use the Deformations toolbox from the "Toolboxes..."
pulldown. Selecting this gives you a little menu on the lower left SPM
window. Select [Deformations]->[Deformations from sn.mat]
Specify the sn.mat file and the default bounding box and voxel size
This will write out a deformation field from the sn.mat file (one image
named y_*.img), which then needs to be inverted.
2) Invert the deformation by [Deformations]->[Invert deformations]. Select
y_*.img as the deformation and then specify your
un-normalised image (this defines bounding box, voxe sizes etc). The result
of this would be an image containing the inverse deformation field (named
iy_*.img).
3) Apply the inverse deformation field to your spatially normalised images
by specifying [Apply deformations]. Then enter the number of subjects, the
deformation field (iy_*.img) and the spatially normalised image.
> 2. And the more important question: by searching the archive, I discovered
> some postings, making me feel doubting, if I am doing the right thing at
> all. It was discussed, that inverting non-linear transformation fields may
> be ambigous, if the mapping was not one-to-one (2 original positions could
> be mapped to one). Since I want to apply the inverted sn_file across
> modalities, I am now a little bit confused. Can I use the SPM spatial
> normalisation routine to implement this at all? Or do I need to set
> constrains to the performance of SPM?
> (http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0208&L=spm&P=R14249&I=-1)
There are no constrains that force the transformations to be one-to-one, but
in practice, because of the low number of parameters and regularisation, they
almost always are.
> 3. I also tried to match the template using the coregistration routine in
> SPM, but I do not know how to invert the resulting mat-file. I also browsed
> the archive for that, but I guess its to trivial. It was mentioned to
> invert it using matlab. Unfortunately, there are 2 matrices in the mat file
> (a M and a mat). Can somebody give hints how to arrange the matrices? And
> what about the image? I need to specify the space defining image; can I
> just leave it untouched and simply replace the accompying mat file?
I would skip this bit.
Best regards,
-John
|