Fmriprep- only calculate the confounds file (or- only calculated estimation of head movements)

Two question:

  1. Which steps in fmriprep “clean” the bold using the confounds calculated (I know that nilearn is using it, but was wondering if the output preprocessed bold is “cleaned” somehow using the confounds calculated)

  2. If one would want to run only the pipeline to produce confounds.tsv files- is there such an option to run only this in fmriprep (reports only?), or I better implement it myself in nipype?


It is not. Temporal denoising needs to be performed outside of FMRIPREP using a strategy of your choosing. There should be other threads on Neurostars about it.

There is no such option. Out of curiosity what use case would it satisfy?

1 Like

Regarding your question:
Estimation of the number of problematic sessions, on a weak computer.

Any reference to such a pipeline (confounds) or example in nipype?

You should check out

1 Like

Thanks, Chris-
I’m familiar with it (will try it soon), but:

  1. Can I run it on a standard desktop (4 cpus, 8 GB RAM) on multiple participants? (fmriprep will never finish on such a machine on multiple participants…)
  2. Does it create a confounds.txt files with head movements?
  3. Any reference for running the counfounds workflow (or any workflow) without fmriprep? (windows docker env)


Yes it should. Remember to enable all the cpus and as much memory as possible in docker settings.

Not quite, but it provides summary values for each run describing motion, outliers, dvars etc. Ideal for assessing quality of scans. See documentation for more details.

1 Like