Troubleshooting: Analysis Topics

Under Construction
Collection of analysis tips/troubleshooting discovered during process. This page is primarily a temporary holding space until these notes can be filed in the proper sections.

Freesurfer - BEM creation

• BOTH FreeSurfer recon-all -autorecon1 and mne.make_watershed_bem independently call mri_watershed to produce a skull-stripped image from which the inner and outer skull surfaces are created.
• The use of -gcaatlas ensures that the bem will pick up the cerebellum, and doesn't clip out other portions of the brain.
• In some cases, -gcaatlas worsens a bem that was already bad but not catastrophic. (i.e. it included some curlicues but didn't crop out any of the brain). This is usually the result of a bad skullstrip, and can be fixed by lowering the default preflood height from 30 to 15, 10, or even 5.
• In very rare cases, -gcaatlas results in a worse skullstrip. (*Observed in subject with Dx.)

Recommended course of action is to use -gcaatlas, but check all bems and change preflood height as necessary.

The use of -gcaatlas does actually improve matters for many subjects, eliminating cropping off of brain structures. Occasionally, using -gcaatlas results in failure, this again is usually a sign that the bem was already weird even without -gcaatlas

For ENIGMA, recommended course of action: 
* Keep -gcaatlas implementation
* Manually run several participants with bem failures and adjust preflood height on mri_watershed/make_watershed_bem
* Inspect output errors for any remaining "no vertices" errors that are potentially due to a bad skullstrip