Dear Pierre-Yves,
There are different aspects, signal drop-out, (spatial) signal distortions, intensity inhomogeneities.
The prescan normalization usually means acquiring additional data resulting in one "homogeneity image", which is then used to scale the images of the functional series - but you can't recover any missing data and you don't account for spatial distortions. Accordingly, the intensity-adjusted images look better overall due to a more homogeneous contrast, but the temporal SNR for individual voxels should remain identical because the standard deviation is up-/downscaled as well.
Nonetheless there can be some benefits when it comes to preprocessing / motion correction. There are some intensity variations tracing back to field inhomogeneities; simplified, e.g. the center is darker no matter where the exact head position is. This will affect motion correction to some extent, as the dark parts in image 1 are mapped onto dark parts of image 2 (leading to an "anchoring effect"). With prescan normalization some of those intensity variations are removed, and with reduced bias the actual head motion should be accounted for to a larger extent during motion correction. To stay on the safe side you could save both raw and prescan-normalized data.
To deal with drop-outs / spatial distortions you would have to additionally adjust slice thickness, orientation of the slices, phase encoding direction, e.g. see http://dx.doi.org/10.1016/j.neuroimage.2004.12.002 , http://dx.doi.org/10.1016/j.neuroimage.2005.08.042 and http://dx.doi.org/10.1371/journal.pone.0008160 . There are several other studies focusing on susceptibility artefacts in amygdala or (oribto)frontal regions which might provide relevant information / ideas for which parameters to modify.
Best
Helmut
|