Dear Marcus,
> can anyone explain the following surprising smoothness estimates:
I will try it , but I'm not quit sure...
> In an effort to assess the overlap between a group level contrast (3x3x3
> mm^3 EPI voxel resolution) and an anatomical map (1x1x1 mm^3 voxel
> resolution) I resliced the single subject contrast images via the SPM2
> GUI ("coregister" & "reslice only" to the colin27 T1 image). Next I
> included the resliced con images in a second level analysis and plotted
> the default SPM glass brain. The t-maps look similar to those obtained
> from the analysis of the original contrast images with 3x3x3 mm^3 EPI
> voxel resolution - of course they are now much less grainy. However the
> glass brain plot footer indicates that the estimated smoothness FWHM is
> 0.1x0.1x0.1 {mm} (see first attached screenshot). - How can this be
> explained?
I don't have a clear explanation, but I could imagine, that this was caused by the fact, that the con-images are not longer masked, when you do some kind of reslicing, i.e. the original con-images should have 'true' values within the brain and NaN outside. You are loosing this masking by doing the reslicing. Have you used the option 'implicite masking'?
Perhaps, you should try both, implicit and explicite masking (with the brainmask of the 'apriori' folder, for example).
For a quick and dirty check, you can calculate the analysed volume out of the number of analysed voxels and the voxel size. It should give you a value, close to the brain volume. Otherwise, you are analysing too much voxels, especially, too much voxels with constant values, which could probably give you a mis-estimated smoothness.
> One other observation that might be connected to the aforementioned is
> that the resliced contrast images are surrounded by a box of "NaN",
> whereas directly at but outside of the brain the voxel values are "0"
> (see second attachment screenshot).
This is caused, I guess, by the fact, that the bounding boxes of the two images are different. Reslicing after coregistration will result in images, which are having the same image properties, as the target image. Since the boundig box of the target image is larger as that one from the con-images, I guess, the rest is filled up with NaN.
Perhaps John could comment here, if I'm completely wrong with my assumptions...
All the best and good luck,
Karsten
--
-------------------------------------------------------------
Karsten Specht, PhD
Department of Biological and Medical Psychology
National Competence Centre for functional MR
University of Bergen
Jonas Lies vei 91
5009 Bergen
Norway
Tel.: +47-555-86279
Fax: +47-555-89872
[log in to unmask]
http://fmri.uib.no/
|