Nuisance covariate regression (head motion, CSF and WM signal, global signal) in local connectivity measures (like Regional Homogeneity (ReHo), Amplitude of Low Frequency Fluctuations (ALFF and fALFF), Degree Centrality and Voxel-Mirrored Homotopic Connectivity (VMHC)) is still a matter of debate. What is your opinion? Is nuisance regression more effective before or after bandpass filtering?
Just my 2 cents:
In general whole brain BOLD signal all have common source of error introduced by physiological noise (CSF & blood flow, pulse, etc.), thus regressing out signals from CSF & WM should increase signal-to-noise ratio. Regressing head motion should in theory reduce error variance at the edges, but if head movements is too vigorous it may not help too much (see http://www.sciencedirect.com/science/article/pii/S1053811905003095).
Whether regressing global signal is still in debate, as it has been shown to cause artificial deactivation for network analysis using seed regions or ICA. However for indices concern regional connectivity it may not matter that much because the calculation is based on coherence/concordance of timeseries.
I would probably perform regression before bandpass filtering because it you do it the other way round you'll temporally smooth your error estimation (CSF & WM signal), which will in turn affect the signal after regression.
There is some merit in throwing out volumes with high motion rather than simply regressing out motion covariates. There have been a number of papers on this topic lately. Here are just a couple.
http://www.sciencedirect.com/science/article/pii/S1053811911008214
http://www.sciencedirect.com/science/article/pii/S1053811911011815
As far as regressing the global signal, most lab tends towards removing it. This article is worth a read on that topic.
http://jn.physiology.org/content/101/6/3270.short
- Badges
- Science topic