Dear Christian
> In a recent study, we tried to run a group comparison between fmri time
> series, acquired after transcranial magnetic stimulation or sham
> stimulation. Stimulation / sham stimulation was applied to the subjects
> on the MRI table and followed by 60 epi scans (11 slices each) over a
> timecourse of 10 minutes. For evaluation I am currently using SPM for
> Windows 95 Version 1.01, which I will refer to as SPM for the sake of
> the argument.
>
> I vaguely remember that, according to Dr. Holmes, using a boxcar
> function, labelling the 60 scans after sham stimulation as baseline,
> and the 60 scans after true TMS as activation, would not work, because
> SPM assumes the time during acquisition of the scans to be continuous.
Strictly speaking this is true but ingnoring temporal correlations this
gaves an appropriate fixed-effects model. If correlations are present
then post hoc temporal smoothing should minimize any effect of
appending different time-series.
> Therefore, I did a group analysis on one subject just to check whether
> this would work at all. Group 1 comprised of the 60 scans after sham
> stimulation and Group 2 of the 60 scans (same subject, different day)
> after true TMS. Since my sample is as small as it could possibly get I
> expected not to find any activation at all. I do however find one
> single cluster (size 5165 voxels) at Z-values of ~9.
>
> Describing this design to the people in the lab did not give me any
> clues as to what I have overlooked. Could anyone point out to the
> mistake I made?
You have made no mistake but you cannot attribute this effect to TMS.
You have used a fixed-effect model to make an inference that requires a
random effects analysis. Although it may be the case that the two
sessions differ this is as meaningful as saying that women are
significantly taller than men on the basis of measuring a tall woman
and a short man 60 times each. To make an inference about TMS you
would need to average the scans from several (say 8) TMS and sham
sessions and then perform a simple group comparison.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|