Mix of fmri with and without in-scanner intensity normalization. Should intensity normalization be applied to all?

I have a dataset containing a mix of fmri scans with and without in-scanner intensity normalization (all data collected on a Siemens scanner). Would it make sense to apply intensity normalization (i.e. N4 via ANTs) to the fmri scans before deconvolution? (that is, either to all scans or only to those without in-scanner norm)

I wonder if the voxelwise scaling of the "scale" block in afni_proc.py would mitigate many of these differences, as long as there is not saturation happening?


Intensity normalization is regularly applied to anatomical data, but it is almost never done with EPI. Very rarely, we apply it to EPI data but only for alignment purposes. The AFNI programs, 3dUnifize and the recent, 3dLocalUnifize, compute intensity normalization.