Dear Torsten,
> I run a data set (2 groups) through randomise (using design_ttest2
> created
> .con and .mat files), and I just want to clarify some of the output.
> The tstat_1 image is the control>patient, and tstat_2 is
> control<patient
> t-statistic.
>
> I assume that this is true, but why would those two images be exact
> mirrors,
> I mean in the first the range is from -1.8 to 18 (approx), and the
> 2nd from
> -18 to 1.8 (at corresponding voxels the signs are just inverted)?
>
I am not really sure what the problem is here. I would actually
expect precisely that, wouldn't you? Think of it is a group mean for
the first group (A) and a group mean for the second group (B), and
then there will be an error estimate SD. You have one of each of
these values for every voxel. So, in your tstat_1 image you will have
values (A-B)/SD, and in tstat_2 (B-A)/SD.
> And why would I get values covering negative and positive values
> within one
> contrast? I thought it could only start at 0 towards either maximum.
Any t-contrast will contain negative and positive values, typically
in roughly equal measures. It is the thresholding that imposes the
non-negativity, i.e. FSL uses a one-tailed test.
I hope that clarifies things.
Good luck Jesper
|