Is/how much fmriprep (freesurfer et al) is resilient to "defacing"?

I know that I could have just tried systematically to compare but decided first to ask.
If we start defacing all our data “by default”, I wonder if

  • fmriprep would produce quantitatively same pre-processed data with dataset carrying defaced anatomicals? Locally some folks did comparison long ago to arrive to the conclusion “the same”, but details on how defaced etc by now are gone
  • what defacing strategy/tool (there is a number of tools) makes least impact
  • may be fmriprep could get an option to use sourcedata/ location to look for anatomicals? That is where e.g. we stored (under git-annex, not shared publicly) original non-defaced ones for some datasets (see e.g.

Thanks in advance for sharing your experience(s)

1 Like

Fmriprep has been developed and tested almost exclusively using defaced data (defaced using different methods). We never saw the need to use non defaced data and I would personally recommend defacing as part of the data curation pipeline for all data coming out of the scanner. It makes it easier to share the data or use it on non secure systems.