Elie,
>With scaling, I observed some significant "deactivations" (C-A
>contrast), that were not observed with Ancova. When reanalyzing the data
>without any normalization, I observed no deactivations, but more
significant activations (A-C
>contrast) than in the Ancova or the Scaling analysis.
The behavior you have observed stems from, as you note, the presence of
correlation between your global signal covariate and your task covariate. As
a result, your global signal covariate is acting as a confound instead of as
a simple nuisance covariate.
I can imagine cases in which one might want to include a confounding
global signal covariate; for example, to explicitly exclude the possibility
that a local signal change can be explained by some effect of global flow.
If included, the interpretation of "activatation" becomes something like:
"those voxels with a relationship with the task
that is significantly greater than that which the
global signal enjoys"
and vice-a-versa for "deactivation". Notably, one cannot infer that
"deactivations" result from decreased signal values during control as
compared to experimental periods.
In most cases, though, my bias is to exclude a confounded global signal
covariate.
You can test for confounding in the future by creating the analysis
model that you intend to use (excluding a global signal covariate), and then
applying that model to the global signal itself. If there is a significant
relationship between your task covariates and the global signal, then the
global signal would be a confounding covariate. I think that this measure
should be reported in any neuroimaging study in which a) correction for the
global signal is undertaken and 2) the authors attempt to interpret
deactivations.
If you can stomach it, further discussion of these topics can be found
in a preprint on our web site:
http://cortex.med.upenn.edu/papers.html
Geoffrey Karl Aguirre, Ph.D.
Univ. of Pennsylvania
(215) 614-1976
mailto:[log in to unmask]
http://cortex.med.upenn.edu/~aguirre
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|