Hi Jeff,

That's a good/complicated question. I would say that in general the normalized vs. raw measures focus on relatively different aspects of the data. In particular, the normalization procedure controls separately each individual image (i.e. separately for each subject and also separately for each condition, e.g. pre/pro), explicitly normalizing the distribution of ALFF values across all voxels so that it is exactly gaussian. The objective of this normalization is to keep ONLY the rank/order of the values across voxels, while disregarding everything else including their absolute scale. Because of this, the end result is that when using 'normalized' ALFF measures you are no longer measuring the absolute amplitude of those fluctuations across subjects, but rather the relative amplitude in different areas (e.g. a positive/negative value in a voxel for a given subject&condition means that this voxel had higher/lower ALFF than other voxels in that same subject&condition).

Beyond differences in interpretation, I imagine that one advantage of using raw (not-normalized) measures would be its ability to observe effects which may be global in nature (i.e. affecting all/most of the brain), while one advantage of using normalized measures is precisely its inherent ability to control for such global effects when those effects are mostly artifactual/confounding (e.g. caused by different acquisition parameters)

Last, in practice if the results of the two analyses differ it may be interesting to follow that up with analyses of global effects (e.g. differences in average/overall BOLD std differences between subjects or conditions) in order to try to determine what may be causing those global differences (e.g. there may be artifactual effects that are causing global differences in ALFF, which may explain negative not-normalized results and positive normalized results; or there may be meaningful global differences in ALFF which may explain negative normalized results but positive not-normalized results). In case it is useful, I have added to the development release the computation of QC_BOLDstd variables separately for each individual condition during the denoising step (these basically compute the average across all voxels of the BOLD signal temporal standard deviation, separately for each individual subject&condition), which may allow you to look in a bit more detail to those global ALFF effects.

Hope this helps
Alfonso

Originally posted by Jeff Browndyke: