Hi Sarina,
> My study uses DTI looking at 4 groups (3 patient and 1
> control). I've run
> TBSS, and even though I get significant voxels appearing
> the fstat image,
> there are no significantly different (corrected) voxels for
> any contrast (I
> tried them all!). The sample sizes of the groups aren't
> very large (and
> slightly uneven) - perhaps this is part of the problem?
I guess that when you say "significant voxels appearing in the fstat image", you mean that you simply see some voxels in this image?
The only "significant" voxels are the corrected ones I'm afraid, so in the *corrp_ftstat1* image above the 0.95 threshold (if you have entered the -F option in randomise).
Sample size can definitely be an issue, but you can always check if you have some results surviving the correction for multiple comparisons with a more lenient threshold for the F-test and with the tfce approach if you haven't used it yet.
> Anyhow, I have
> started looking at interhemispheric differences in FA
> within each group.
> However, I am unsure as to how exactly I would go about
> comparing
> asymmetries statistically between the groups using TBSS.
>
> Could I run separate tbss_symm for each group and find
> R>L or L>R
> differences? Although I would probably come across the
> problem of how to
> compare between groups.
>
> Anyone have any ideas/suggestions? Would greatly appreciate
> it.
You just need to run tbss_sym FA in your stats directory and then ask your different questions in the design model:
1 0 0 0 where is L>R in my first group
-1 0 0 0 where is R>L in my first group
0 1 0 0 where is L>R in my second group
0 -1 0 0 etc.
1 -1 0 0 where is L>R in my first group superior to L>R in my second group. Bear in mind that this is equivalent in the GLM to: where is R>L in my second group superior to R>L in my first group, hence the need for contrasts in each group to disentangle these two interpretations!
-1 1 0 0 etc.
Hope this helps,
Gwenaelle
|