Mean scaling

Mean scaling is a GLM option that allows you to normalize the mean signal from each voxel in each run, dividing by the mean so that the mean is 1. It removes scaling differences between runs (what you'd expect if your signal were multiplied by a different value for each run, not level differences (what you'd expect if a different constant were added to your signal from each run). The latter can be modeled out with scan effect covariates.

Mean scaling, like Linear detrending, is not applied to your image data files, it is applied to your data internally just before regression. It is also an option in viewing your data.