To me, it looks like an issue with the generation of the mask used for
determining which regions to include when fitting the GLM. I suggest
you use the Check Reg button to see how well your spatially normalised
images are aligned with the images that are supplied with SPM.
Also make sure that the fMRI and anatomical data are well aligned
together (Check Reg) before you do any spatially normalising based on
the anatomical.
The best procedure varies from scanner to scanner. If you can achieve
good alignment between fMRI and anatomical, then the segmentation
approach should do a reasonable job. If accurate coregistration is not
possible because of distortions (which could maybe be corrected via the
FieldMap toolbox), then a different approach may be preferable.
Sometimes it is possible to segment the fMRI directly and use the
resulting deformation to spatially normalise the data.
Personally though, I would use the DARTEL toolbox of SPM8 to do my
spatial normalisation. Instructions are in the spm8/man/manual.pdf
file.
Best regards,
-John
On Wed, 2009-08-12 at 14:19 -0400, Chris Watson wrote:
> Hello,
> I was wondering if anyone has seen something similar to what I'm going
> to describe. I'm attaching a screenshot of results from a finger-tapping
> experiment.
> As can be seen, the edge of the blobs are several millimeters away from
> the edge of the template image. Looking at the mask image and the
> template together also show this. I have seen this with almost all of
> the subjects. Could it be the subjects' age (mean age 11-12 years)? I
> ran a couple of subjects using the regular Normalization step and the
> mean functional image as the source, and it looks better, but I haven't
> done all the subjects yet. I considered making a study-specific
> template, but unfortunately n=8 for both groups (I'm assuming there are
> too few subjects). I don't expect perfect registration but there are
> several millimeters difference in some areas, as if the warped images
> are "shrunken" compared to the template.
>
> So, even though Unified Segmentation is the suggested method, would it
> be invalid to use the Normalization step?
>
> Thanks,
> Chris
--
John Ashburner <[log in to unmask]>
|