bem.make_watershed_bem with gcatlas fails on default Freesurfer 8 processing

[pmolfese]

Description of the problem

It appears that Freesurfer 8's shift to synthstrip by default no longer generates the talairach_with_skull.lta file necessary for bem.make_watershed_bem to run with gcaatlas=True. The missing file can be created by running the following command.

mri_em_register -skull nu.mgz /usr/local/apps/freesurfer/7.4.1/average/RB_all_withskull_2020_01_02.gca transforms/talairach_with_skull.lta

Not sure whether we should add this step as a check in the function and run before the mri_watershed or perhaps just check if the file exists and then advise user to run the command separately on failure.

Steps to reproduce

1. Run Freesurfer v.8.0: recon-all -s {subj} -I {dicom_path} -all
2. attempt to run watershed from python with gcaatlas:
   mne.bem.make_watershed_bem(subject, sourcepath, overwrite=True, atlas=True, gcaatlas=True, show=True)
3. Note missing file: {subj}/mri/transforms/talairach_with_skull.lta

Link to data

No response

Expected results

Function mne.bem.make_watershed_bem when given gcatlas=True would run mri_watershed and create the BEM.

Actual results

Function mne.bem.make_watershed_bem when given gcatlas=True fails.

Additional information

I can probably do a PR to fix, just curious how everyone feels about 1. should we run the mri_em_register for the user if file is missing or 2. error out and advise them to run command?

[larsoner]

Hmmm... I think my vote would be to try to generate the file if it's missing, and add a note to gcaatlas param in the docstring saying that with FreeSurfer 8+ we will try to generate the necessary file. @pmolfese you up for making a PR to fix it?

[pmolfese]

I'll give it a shot this week!