External Email - Use Caution

Hello,

I am working with a longitudinal dataset that was initially processed entirely with FreeSurfer 8.0.0.

The cross-sectional time points were manually edited and then rerun with FreeSurfer 8.0.0. The base subjects and longitudinal subjects were also created with 8.0.0. I am now at the stage where I may need to manually edit the base subjects and possibly the longitudinal subjects, and I am unsure which FreeSurfer version I should use for the remaining edits and reruns.

My concern is the documented FreeSurfer 8.0.0 bug in which rerunning recon-all can turn mri/orig.mgz into a symbolic link to mri/nu.mgz, particularly after manual editing and rerunning the processing. As far as I can tell, this bug did not occur in my dataset. I checked all relevant directories recursively and found no orig.mgz symbolic links and no cases in which orig.mgz and nu.mgz were byte-for-byte identical.

My questions are:

1. Since the cross-sectional subjects were already manually edited and rerun with FreeSurfer 8.0.0, should I continue using 8.0.0 for manual edits and reruns of the base and longitudinal subjects in order to keep the processing version consistent, while checking that the symbolic-link error does not occur?

2. Alternatively, since it would take too much time to do all the manual edits all over again, can I keep the existing cross-sectional subjects and switch to a newer FreeSurfer version for the base and longitudinal processing?

3. If I should switch, would you recommend FreeSurfer 8.1.0 or 8.2.0?

4. Would it be acceptable to keep the existing base subjects created with 8.0.0 and use FreeSurfer 8.1.0 or 8.2.0 only for editing and rerunning the base and longitudinal subjects? Or should I first recreate the base subjects with 8.1.0 or 8.2.0 and only then begin the manual editing?

I also have a separate question about the longitudinal results. Is it expected or possible that:

* the cross-sectional subject looks correct after editing and rerunning,

* the base subject also looks correct,

* but the corresponding longitudinal subject contains pial-surface errors that are not present in either the cross-sectional or base subject? The longitudinal subjects look worse than the edited and rerun cross-sectional subjects and also worse than the base subjects.

The study period was about one year, with two to three MRI sessions per participant. I observed this both in adolescent participants with anorexia nervosa who gained weight over the one-year period (and in whom substantial brain volume changes would be expected), and in adolescent healthy controls (in whom normal developmental brain changes are also expected). Could it indicate that something went wrong during the creation of the longitudinal subjects? If so, what is the most likely cause?

Thank you very much for your help!

Best regards,

Andreas