Synchronous temporal filter workflow

[rciric]

I have a workflow that applies the same temporal filtering procedure to both 4D NIfTI time series and 2D TSV time series (e.g., confounds). This is necessary to avoid frequency mismatch issues that can potentially reintroduce nuisance signals at particular frequencies during the regression/fit step.

I would like to make this workflow available, but I'm not sure what the best place for this is (niworkflows vs. niflows vs. fmriprep). In brief:

• This would be a dependency of the Power denoising workflow that I was hoping to integrate into fmriprep down the line.
• The workflow could also potentially be a standalone for synchronous filtering of fmriprep outputs before feeding them into fitlins.
• If we build a separate confound regression/denoising/qcfc module, this would also be an explicit dependency for that.
• For users who are are all right using an ideal filter/cosine basis, this workflow is generally unnecessary. I haven't really investigated the merits of the different filtering approaches, but both the cosine basis method (simultaneous filtering) and the parallel generalised filtering method implemented in this workflow should in theory get around the frequency mismatch/nuisance reintroduction problem.

[kfinc]

I'm wondering whether it would be better to incorporate synchronous filtering option into fitlins, similarly as it was done with smoothing. Choice of both smoothing and filtering parameters depends on the type of analysis you want to perform (i.e., it may be different for activation analysis, functional connectivity analysis, RSA), so I think it is fine to keep it outside preprocessing (as standalone workflow or/and within fitlins).

Have you looked into a new version of nilearn.image.clean_img? It incorporates synchronous temporal filtering, according to Lindquist et al. (2018) recommendation.

[oesteban]

WDYT @effigies, should we redirect this initiative to fitlins?

@rciric is there any element from fMRIPrep that will make it difficult including this within fitlins? Where would you prefer to see this included?

[effigies]

I don't think it'd be wrong to have in FitLins, but bear in mind that FitLins is more-or-less GLM-specific, and it might make more sense to have this filtering as a separate pipeline that can be used to feed into resting state and other, non-GLM analyses. If that exists as a Nipype workflow, it shouldn't be too hard for me to slide it into FitLins.

[oesteban]

Let's add the workflow in niworkflows. We should also check how much overlap is there w.r.t. nilearn.image.clean_img.