Hi,
I am trying to do a general functional connectivity analysis as described in:
Elliott, Maxwell L., et al. “General functional connectivity: Shared features of resting-state and task fMRI drive reliable and heritable individual differences in functional brain networks.” NeuroImage 189 (2019): 516-532.
However, I had questions when I was doing 3dDeconvolve and bandpassing. I have seen two methods, one generating bandpass regressors via 1dBport and using these regressors in 3dDeconvolve, and then running anaticor, as below:
“Method 1”
3dDeconvolve -input pb00.$subj.r*.tcat+tlrc.HEAD
-censor motion_${subj}_censor.1D
-ortvec bandpass_rall.1D bandpass
-polort A -float
The other method (“Method 2”) involves using 3dDeconvolve without the bandpass regressors, running anaticor, and then using 3drsfc with nodetrend, as below:
3dRSFC -prefix gfcanalysis_output -nodetrend 0.01 0.1 errts.test.$subj.fanaticor+tlrc
Using these two analyses steps results in somewhat different results that appear to cover similar regions when subsequently calculating seed-based RSFC, but are certainly not identical. Is there rationale for choosing one method over the other? I thought that doing Method 1 might be more sound given that there is only one regression step, but I am unsure and don’t have too much experience in conducting RSFC analyses.
Any help would be greatly appreciated, thank you !!
Bandpassing should be included within the regression modeling. Doing it in 2 steps is basically incorrect mathematically; unfortunately, it seems a fair number of studies use this latter case. One risk is that you can use up all your available degrees of freedom without knowing it (and hence basically reducing your data to noise). Another risk is that when processing in 2 steps like that, you can remove some signals in the first step and then (unwantedly) put them back in the second. The AFNI FMRI processing setup tool, afni_proc.py, helps you do this correctly. (NB: mathematically, bandpassing can be accomplished either with Fourier based methods or with a regression model-- they can be viewed as equivalent strategies. The latter can be integrated with regression modeling, which is what we recommend.)
This is described more here: https://afni.nimh.nih.gov/pub/dist/doc/htmldoc/programs/afni_proc.py_sphx.html#resting-state-note
See also the comments there on taking time to be sure that you doooo want to bandpass- is it necessary+desirable, or is it being done just because other papers do? Bandpassing RS-FMRI data often drastically decreases the number of degrees of freedom available for further analyses.
See also this paper: https://pubmed.ncbi.nlm.nih.gov/23747457/
→ The nuisance of nuisance regression: spectral misspecification in a common approach to resting-state fMRI preprocessing reintroduces noise and obscures functional connectivity
Michael N Hallquist, Kai Hwang, Beatriz Luna
–pt
ps: it iiiis possible to bandpass in 2 separate steps in a mathematically correct way, if you bandpass your regressors before using them in the regression model. That is:
bandpass your regressors
bandpass your time series
make a GLM of your bandpassed time series with your bandpassed regressors.
However, this method is susceptible to mistakes, I think, and it is much cleaner to make one regression model.
Thank you, this is very clear and a great explanation - will definitely take a deeper dive into those manuscripts. Greatly appreciate it!
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.