Chetan,
To co-register from PET to MRI, should only
be a fixed body problem (Rotations and Translations).
The scaling should be correct from the scanners. If
not, there is an issue with the files being created from
the scanners, possibly while converting formats.
The normalization process, typically, isn't just a
linear transformation but also includes non-linear
components. The MRIs have higher resolution,
and we like to take advantage of this when we do the
normalization, which has more parameters to it than
a fixed body problem.
Also, you're not done with the normalization, until
you check the results, at a minimum with "check reg".
We have found that the MRIs give us a better normalization,
than we could have achieved using PET alone.
For you last point, we don't co-register from the PET to the MRI template
for a couple of reasons.
1) We co-register when the images should be of the same size and
shape, only mis-aligned. Ie, you coregister from the same person
PET to MRI. You normalize (not coregister) to go
from the "subject" space to that of the template MR.
2) When you do normalize, you would use the MRI template
to normalize MRIs, and the PET template to normalize "water"
PET scans. For other PET tracers, you don't have a template to normalize
to unless you create one.
Talk to me this week, and I'll work with you on this.
david wack
university at buffalo
Chetan Bhole wrote:
> Hi,
>
> We have been for quite some time following process:
>
> 1. Co-register the PET patient images to the MRI patient images to get
> the transformation matrix T1
> 2. Normalize the MRI patient image to the MRI template to get
> transformation matrix T2.
> 3. Can change the PET images using T1*T2 to coregister to MRI template
> image.
>
> Why is it not a good idea to coregister directly from the patient PET
> images to the MRI template image.
>
> The MRI patient image may have come from a different machine (that the
> PET machine - those that have both PET and MRI) and the MRI machines
> might have different resolutions, scaling factors, shear variables
> than the PET machines. So it is not like we only may need to rotate
> and translate the PET image to the MRI patient image. We may also need
> to scale and shear it. So why not directly normalize
> (rotate,translate, shear, scale) directly with respect to the MRI
> template?
>
>
> Thanks and Regards,
> Chetan
|