open-discussion > When to use global signal correction
Showing 1-1 of 1 posts
Results per page:
Oct 17, 2008  04:10 PM | Daniel Handwerker
When to use global signal correction
Since this is supposed to be a discussion list, I figured I'd try to start a discussion. I'm going to be a bit more dogmatic and provocative here than we were in the article since I figure the best way to start an active discussion here is to stake a position that I think others might want to disagree with. Please feel free to invite others who might have an interest in this topic to join the discussion.

We have a Neuroimage article in press:
The Impact of Global Signal Regression on Resting State Correlations: Are Anti-Correlated Networks Introduced?
Available online 11 October 2008
Kevin Murphy, Rasmus M. Birn, Daniel A. Handwerker, Tyler B. Jones, Peter A. Bandettini
If you want a copy of the article and don't have access, you can email me or one of the other authors.

In the article, we show with both theory and data how global signal regression will create anti-correlations in resting connectivity maps even when no such anti-correlations exist. We do not say that all anti-correlations following global signal regression are spurious.

1. Still, given, that we now know artificial anti-correlations will be created, is it ever appropriate regress out the global signal before running a seed-based correlation analysis?

2. If an anti-correlated network appears after global signal regression, but not after other methods that try to remove physiological noise, can we trust that the anti-correlated network has neural underpinnings? If not, is there strong evidence beyond fMRI connectivity studies using global signal regression that the default mode network has correlated and anti-correlated regions with a neural origin? The last few paragraphs of the discussion in the manuscript outline the connectivity results using some other analysis methods. I'll again note that I'm intentionally being a bit provocative here.

3. What methods should we use to remove non-neural noise in connectivity studies. In the article we use RETROICOR and Respiration*Volume/Time (RVT) correction, but there are other methods that just regress out CSF or white matter signals. What do the people in this discussion group consider their standard method?