> Newbie with silly questions here. I'm running a SPM8 VBM study on sleep disturbances comparing a group of MRIs from 31 insomniacs vs. 113 subjects with normal sleep. All non-demented. I have more or less followed the steps of Dr. Ashburner tutorial:
> reorient-> -segment-> create a single template for the 144 subjects -> smooth.
> For the analysis we have run a two sample t-test controlling for age and total volume (groups were not different in gender). We have found some differences between insomniacs and normals but we wonder if this differences could be explained by the different size of each group. Should we use two groups of 31 subjects instead?
I don't think that the differences could be explained by the
differences in the number of subjects per se. Having more subjects in
a group should give you a more accurate estimate of the mean gray
matter volume (GMV) and how variable it is across the group. However,
a few other thoughts come to mind:
1) It appears you've used DARTEL to create a template for all of your
144 subjects. This template will be biased towards the controls,
because there are many more controls than insomniacs. If there are
indeed shape (i.e., GMV) differences between the groups, then this
might mean controls must be warped less to match the template, and
insomniacs warped more (because the template is more similar to the
controls' brains). This may influence your results. Thus, I would
suggest creating a template that is matched between controls and
insomniacs. You may want to select 31 controls that match the
insomniacs in age and sex, for example. Create a template based on
these 62 subjects, and then warp all 144 subjects to this (unbiased)
template. At least, that would be my inclinationóbut I would be
grateful to hear others' opinions on this issue.
2) If variables like age or total GMV vary smoothly across all of your
subjects, then adding them in as covariates will likely explain some
variance and potentially make your results easier to interpret.
However, if your groups differ significantly in these variables (i.e.
insomnia diagnosis is not independent of age or total GMV), then
including them may affect the results you are interested in. It may
be ok, in that your group comparison would be conservative (i.e.,
looking at results for insomnia only reveals regions that can uniquely
be explained by insomnia diagnosis). However, it's just something to
keep in mind as you interpret your results.
> Also, should we create a new template for this analysis or can we use the template that we used for the former study with the new one? Or is it better to use a pre-existing template?
The idea behind using a custom template is that for any sample of
brains, they are likely to be closer in shape to their average shape
(a custom template) than some standard template. If they are closer
in shape, the registration should be easier, and they should be more
closely aligned. So, I think you will get more accurate results to
use custom templates for each analysis, as described in the SPM manual
and John Ashburner's DARTEL manual (taking into account the issue of
equal weighting discussed above).
Hope this helps!
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104