If you are truly interested in the effect of a treatment on a particular
subject (i.e. because you are interested in making a treatment decision rather
than testing a hypothesis), then you will have to get creative. Typically one
would do dual regression and then use between-subject variance in a t-test or
paired t-test. In your case, there is only one subject so the voxelwise
variance has to come from somewhere else.
You could, at least in principle, run MELODIC with temporal concatenation and
then use stage 2 from dual regression to get the "mean" coactivation for each
voxel before and after treatment. To do a t-test, you would then need to
somehow come up with an estimate of the noise for each voxel -- a measure of
within-subject voxelwise variance. I suppose you could use could use the
voxelwise standard deviation from the raw time courses before and after
treatment as a surrogate. You might scale these values by the ratio of the
voxelwise "mean" from dual regression to the raw voxelwise mean.
On Friday, April 06, 2012 16:49:11 you wrote:
> Thanks Benjamin for the answer
> OK, the question for me is still if any inference can be made on a single
> patient; that is the ultimate goal of understanding if a treatment had an
> effect on a case-by-case approach. The group analysis certainly makes
> sense for aggregate results but for individual assessments. Perhaps the
> way to do it is to repeat measures before/after treatment, concatenate
> them and feed them to the analysis? Lino
|