Skull-stripping failure on Quickshear-defaced images?

[effigies]

What version of fMRIPrep are you using?

20.2.0rc0

What kind of installation are you using? Containers (Singularity, Docker), or "bare-metal"?

TACC run

Details

@jbwexler reported that ds001105 gets stuck on brain_extraction_wf.init_aff for all subjects. Here is an example T1w:

https://openneuro.org/datasets/ds001105/file-display/sub-006:anat:sub-006_T1w.nii.gz

This is clearly a Quickshear-defaced image. It's possible this is the source of the problem? I think the majority of OpenNeuro is defaced with pydeface, so it wouldn't be out of the realm of possibility that we've never worked with a quickshear image.

[effigies]

Several other datasets I've reviewed today are quicksheared, so it's not purely that.

[oesteban]

What's the dtype of these images? Could it be a memory issue?

[effigies]

It's a T1w image...

[oesteban]

yup, but ITK will definitely read float64 and keep that width.

[effigies]

nib-ls:

ds001105/sub-006/anat/sub-006_T1w.nii.gz float32 [192, 256, 256] 0.90x0.94x0.94

Nothing extraordinary.