> how would one then coregister images of different sizes? At least some
> zooming should be involved there, be it estimated or simply calculated
> from the headers. I was referring to a slide from your presentation:
> http://www.fil.ion.ucl.ac.uk/spm/course/slides02/spatial/img8.htm
Zooms are computed from the voxel sizes in the image headers (and possibly
.mat files). Most scanners fill this information in correctly (or
approximately correctly), and the conversion routines you use should take
these voxel sizes (and various slice positioning bits of info) into account.
If the scanners get it wrong, then the parameters in the headers should be
changed to correct this.
If all has gone well, then a six-parameter rigid registration should be enough
to correctly register images of the same subject.
>
> OK, this is for spm99, and now many people use mutual info
> coregistration instead of partitioning into GM/WM using templates, the
> latter might be outdated by now. When I investigate the matrices in the
> mat files for spm2 there is zooming indeed, but that probably reflects
> voxel-to-world mappings, isn't it?
It depends how the transform is represented. In SPM, I assume a single
voxel-to-world mapping for each image file. This encodes the orientation,
position, voxel sizes etc, and is a general purpose affien transformation.
After coregistration, the image that is assumed to have moved has its
voxel-to-world transform pre-multiplied by a rigid-body transform.
>
> And the coregistration step (the linear part) in normalization does do a
> full affine transformation, as far as I understood? Correct me if I'm
> wrong...
The initial affine part of the spatial normalisation estimates a full
12-parameter affine transform matrix.
Best regards,
-John
|