fMRIPrep aCompCor, tCompCor, Cosine and highpass filter at 180s

Software Support
,
1

Hi,

Like Milena (see fMRIPrep CompCor and cosine regressors: Include together with high-pass filtering in SPM12?), I am struggling with suggestions from fMRIPrep to include CompCor and cosine regressors in my SPM model. However, the solution to that question does not apply here.

I have a slow task design, so that I need to increase the highpass filter to 180s to not filter out slow signal. fMRIPrep instead uses a highpass filter of 128s for creating the CompCor regressors. I read that there is no option in fMRIPrep (yet) to change the cut off value.

Therefore, I would like to know whether I should:

  • (1) do not include the CompCor and cosine regressors at all and just apply a highpass filter of 180s in SPM? - perhaps use other advanced motion regressors instead, like ICA-AROMA?
  • (2) apply a highpass filter of 180s in SPM, do not include the cosine regressors, but still include the CompCor regressors?
  • (3) use another solution?

(This post is also similar to the unresolved question: Filtering BOLD signal in AFNI)

Thanks so much for any feedback!

Best,
Natalie

2

Hi @Natalie, and welcome to neurostars!

I would avoid option (2), because it creates a mismatch: fMRIPrep’s aCompCor components are computed after applying a 128 s high-pass (via cosine bases), whereas your SPM model would impose a different (180 s) filter. That inconsistency can leave residual low-frequency structure in the data/regressors.

In principle, you could recompute aCompCor after applying your desired high-pass cutoff, but if you want to keep things simple, it may be easier to omit aCompCor and use a different nuisance strategy if you truly need the 180 s cutoff.

A reasonable alternative is to include regressors indexing physiological noise, such as the average CSF signal (optionally with derivatives and quadratic expansions, if you have sufficient temporal degrees of freedom). While this is not identical to aCompCor, it can capture similar sources of variance.

Best,
Steven

3

Hi @Steven, thanks a lot for this clear and quick reply!