Dear Joram,
Global signal is always computed scan-by-scan and stored in SPM.xGX.rg.
The actual scaling of the data, derived from the global above, is stored
in SPM.xGX.gSF and will differ depending on whether session-specific
grand mean scaling is used or not.
This is implemented here:
https://github.com/spm/spm12/blob/r7771/spm_fmri_spm_ui.m#L300-L354
Note that a rescaling of the data (i.e. multiplicative effect) is
applied, as opposed to the use of a confounding covariate (i.e. additive
effect). These considerations were already discussed in the very early
days of SPM for PET data:
https://www.fil.ion.ucl.ac.uk/spm/doc/biblio/Year/1990.html
Best regards,
Guillaume.
On 31/01/2020 10:02, Joram Soch wrote:
> Dear all,
>
> the module "fMRI design specification" has a parameter called "Global normalisation" with options "None" (the default) and "Scaling". If I understand it correctly, if set to "Scaling", the scan-wise global signal across all in-mask voxels is calculated and regressed out of the BOLD signal before/during (?) GLM estimation.
>
> I would like to know
> 1) where in the "SPM.mat" information about this procedure is being saved? (There is a field SPM.xGX, but it seems to exist regardless of whether "Global normalisation" is set to "None" or "Scaling" [which is stored into SPM.xGX.iGXcalc].)
> 2) how this information is used during SPM model estimation? (Does it work the same way as with the temporal drifts (SPM.xX.K), i.e. regressing out the nuisance regressors (SPM.xX.K(i).X0) and taking the residuals for subsequent analyses? If yes, where in "spm_spm.m" do I find the code for this? If no, in which other way is it done?)
>
> Cheers
> Joram
>
> _______________________________________
>
> Joram Soch
> Research Associate
>
> Bernstein Center for Computational Neuroscience
> Humboldt-Universität zu Berlin
> Philippstraße 13, Haus 6
> 10115 Berlin, Germany
>
> Phone: +49 (0)30 2093 6766
> Mobile: +49 (0)176 2830 3980
> E-Mail: [log in to unmask]
>
> Most recent paper: "model assessment, comparison and selection"
> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.sciencedirect.com%2Fscience%2Farticle%2Fpii%2FS0165027018301468&data=02%7C01%7C%7C62cab48eebe84115ff1208d7a634af35%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637160617515943316&sdata=k5jM%2BcOA8bMrI%2BsHBNU%2FS922Iu%2FrcqtdsB5T0AM4wsk%3D&reserved=0
>
> Most recent preprint: "inverse transformed encoding models"
> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.biorxiv.org%2Fcontent%2F10.1101%2F610626v3&data=02%7C01%7C%7C62cab48eebe84115ff1208d7a634af35%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637160617515943316&sdata=ISgMDGbQmaDgE3LV5V%2BEi%2FXQwguuFcw1JeYn8jq%2BPbY%3D&reserved=0
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|