Confounds regression for fALFF calculation

AFNI Message Board
1

AFNI version info (afni -ver): Version AFNI_99.99.99 'Alexios V Doukas the Bushy-eyebrowed'

Hi everyone!

I am working on calculating fALFF and ReHo values for a large dataset. My data was preprocessed with fMRIprep (23.0.2). My plan is to use the 3dRSFC function to extract all the LFF related measures, using the -ort fllag for counfound regression, and then 3dReHo. I was wondering which would be the recommended confounds to include for regression (i.e. what to include in the 1D file provided for -ort). I've been searching but found out benchmarking papers are usually focused directly on how to pre-process data for functional connectivity analysis (for example Ciric et al 2017), but there is not so much information about how these decisions affect LFF metrics. I was thinking of adding just the 6 basic motion parameters, CSF and WM.
Additionally, I'd like to know if it would be advisable to use the LFF 4D volume for later connectivity analyses or if it would be more advisable to reprocess it with it's own confounds and only highpass filtering (instead of the bandpass filtering used for ALFF/fALFF).

Here is how my code looks like for now:

    ## Remove the first 5 volumes from the preprocessed BOLD data
    preproc_bold=${func_dir}${sub}_${ses}_task-rest_space-MNI152NLin2009cAsym_res-2_desc-preproc_bold.nii.gz
    3dTcat -prefix ${tmpsubdir}/${sub}_bold_tcat $preproc_bold'[5..$]' 
 
    ## Run 3dRSFC to get fALFF, ALFF, mALFF, RSFA
    3dRSFC 	-mask ${brain_mask} \
    -prefix ${tmpsubdir}/${sub}_${ses} \
    -ort ${afni_confounds} \
    -despike \
    -quiet \
    0.01 0.08 ${tmpsubdir}/${sub}_bold_tcat+tlrc 
    
    3dReHo  -prefix ${tmpsubdir}/${sub}_${ses}_ReHo \
    -inset ${tmpsubdir}/${sub}_${ses}_LFF+tlrc \
    -mask ${brain_mask}

Thank you very much!
Aliza

2

Hi-

Sorry I missed this earlier, and will reply here. (I see you have also posted the same question here, since we didn't reply--apologies again.)

Yes, you will still have to do a few processing steps, such as blurring (if doing voxelwise analysis, which it sounds like is the case), potentially scaling the time series (to have meaningful BOLD % signal change interpretation), and then performing the heart of the analysis, regression modeling.

In terms of what we would typically recommend for typical resting state analysis, we have a description here:

  • Reynolds RC, Glen DR, Chen G, Saad ZS, Cox RW, Taylor PA (2024). Processing, evaluating and understanding FMRI data with afni_proc.py. Imaging Neuroscience 2:1-52. https://doi.org/10.1162/imag_a_00347

You could probably put your results into afni_proc.py to complete your processing and perform the regression modeling therein. As a benefit, you would have some aspects of the automatically generated APQC HTML to check some quality control features.

To the processing: you will have a question of whether your nuisance regressors include censoring time points or not. That is a common thing to include in resting state analysis. At that point, estimating RSFC paramaters like fALFF becomes more complicated, because it is calculated from frequency spectra, and performing the Fourier Transform assumes that the time points are uniformly sampled; censoring breaks this assumption. The current way in AFNI to still estimate fALFF (or associated ALFF) would be to use 3dLombScargle followed by 3dAmpToRSFC. This is described a bit more in this post, too.

Also, I think from the AFNI version you cite, you are using NeuroDocker. (That AFNI version number of about AFNI_99.99.99 'Alexios V Doukas the Bushy-eyebrowed' signifies that.) I'm not sure what version of AFNI that really is, as that is not a version number we give but one they relabel to, and it never changes. It would be good to know, as we have recently updated 3dReHo, to make sure you have the most recent version.

--pt

3

Hi!
Thank you for your answer, and sorry for now taking forever to respond. I've been going over the afni_proc.py paper and documentation and have some follow-up questions.
For starters, I was originally not censoring the time-series preciselly because RSFC parameters require the whole spectrum. My current understanding is that the way to go to calculate fALFF would be to:

  1. blur and scale the time series
  2. perform regression*, including bandpassing and censoring
  3. use 3dLombScargle (followed by 3dAmpToRSFC)
    * regressing out the six basic motion parameters with their corresponding first derivatives (12 params), and n-first components of acompcor

Now, I figure it might be worth processing the time series independently for ReHo and functional connectivity analyses by regressing out the same components as in the previous case, censoring, but using only a high-pass filter instead of a bandpass filter (following the discussion in appendix A of the afni_proc.py paper). Would that seem reasonable?

For the afni version, I'm using a version installed in an HCP cluser, so I figure it might come from neurodocker, however, I think it is afni/26.0.09

Again thank you very much for your time,
Aliza