inconsistent defacing results

[noah10]

We're finding inconsistent results in defacing depending on the hardware and minor dependency version changes. In the image below, the red is:

• pydeface 2.0.2
• fsl 6.0.4
• python 3.10.4
• OS Ventura 13.2.1 on Intel Mac

...and the green is:

• pydeface 2.0.2
• fsl 6.0.5.1
• python 3.10.1
• OS debian:bullseye-slim (run in Docker with --platform=linux/x86_64 on an M1 Mac, fwiw)

The green is what we see with pydeface 2.0.0 and 2.0.2 on fsl 5 and 6 on a variety of python/debian version combinations; so far it's only the Mac that's different. Unfortunately, the green is also obviously not very well defaced. (The red has its own defacing problems, of course, just different ones.) The original nifti is a normalized PDw high-resolution scan targeting the locus coeruleus.

Does anyone have any thoughts as to why we're seeing different results on different hardware platforms?

[ofgulban]

Hi @noah10 ,

pydeface is basically a wrapper for fsl flirt registration program. See lines:

pydeface/pydeface/utils.py

Lines 90 to 112 in 69a17a6

	print('Defacing...\n %s' % infile)
	# register template to infile
	outfile_type = get_outfile_type(template_reg)
	flirt = fsl.FLIRT()
	flirt.inputs.cost_func = cost
	flirt.inputs.in_file = template
	flirt.inputs.out_matrix_file = template_reg_mat
	flirt.inputs.out_file = template_reg
	flirt.inputs.output_type = outfile_type
	flirt.inputs.reference = infile
	flirt.run()
	outfile_type = get_outfile_type(warped_mask)
	# warp facemask to infile
	flirt = fsl.FLIRT()
	flirt.inputs.in_file = facemask
	flirt.inputs.in_matrix_file = template_reg_mat
	flirt.inputs.apply_xfm = True
	flirt.inputs.reference = infile
	flirt.inputs.out_file = warped_mask
	flirt.inputs.output_type = outfile_type
	flirt.inputs.out_matrix_file = warped_mask_mat
	flirt.run()

Therefore, I suspect that the difference you are seeing is related to fsl version differences and its dependencies.