On the relation between Global Signal Normalization, Global Signal Subtraction, and Global Signal Regression in Resting-State fMRI

In resting-state fMRI, global signal regression (GSR), global signal subtraction (GSS) and global signal normalization (GSN) are widely used nuisance removal methods. So far these techniques have been treated as distinct operations and the relation between them has not been clearly described. In this paper, we mathematically and empirically show that GSS and GSN are nearly identical processes in resting-state fMRI. We further show that in terms of resting-state functional connectivity maps, GSS and hence GSN are similar processes to GSR when considering seed time courses that have a good fit to the global signal time course.