Hi Chris,
On Fri, Jul 15, 2011 at 7:54 PM, Chris Filo Gorgolewski
<[log in to unmask]> wrote:
> Dear SPMers,
> As far as I know SPM provides two methods of removing global signal
> effects: scaling with scan (volume) mean or session mean. I managed to
> find the code responsible for calculating those means and applying
> corrections, but I was wondering if those means were later on used as
> nuisance regressors (and if yes where is it in the code). I also heard
> heard that SPM was departing from removing global signal due to
> negative activation one can get. Therefore it would be great if
> someone could shed some light on this issues.
I've not paid close attention to this recently, but here's my understanding:
Each scan i=1..N has a global value g_i, giving a vector g of globals, length N.
The default in FMRI is to normalize the average global signal in the
run to 100, that is, each voxel value in scan i is multiplied by g_i /
mean(g) * 100 - for each session separately.
The default in PET is to add the vector g as a covariate of no
interest (ANCOVA adjustment).
There was a phase were we were dividing each voxel value by g_i -
proportional scaling. This was the thing generating most of the odd
negative signal, and it's very rare to do that these days.
I think that's right, please y'all do correct me if I'm wrong.
Best,
Matthew
|