Fmriprep: skull stripped inputs not auto-detected for ds001110 [sherlockmerlin]

[jsmentch]

version 20.2.0, using singularity.

The sherlock merlin dataset (openneuro ds001110) has images that were already skull-stripped and fmriprep did not auto-detect that they were skull stripped (this seems to have happened for all of the subjects, not just one or a few). Passing the flag '--skull-strip-t1w skip' did indeed skip skull stripping as expected though.

report after running fmriprep naively with no option set for --skull-strip-t1w (default 'auto'):

running with '--skull-strip-t1w skip':

[mgxd]

Hi @jsmentch - the default option is force.

https://github.com/nipreps/fmriprep/blob/67e085c00f1561188a18c79d3e328d2ccaaf063e/fmriprep/cli/parser.py#L394-L398

[jsmentch]

@mgxd ah, well that resolves this then... thank you!

[effigies]

Can you report back on the success of auto? Part of the reason for defaulting to force was lack of confidence in the heuristic.

[jsmentch]

Hi @effigies I've rerun this with auto and things look good for the most part, however, it appears that sub-27 was re-skull stripped, see image:

[effigies]

Ah, thanks! Good opportunity to improve our heuristic. I'm going to reopen this in smriprep where the heuristic is implemented.