Relevance of Freesurfer expert settings for MNE workflow

✉️ Mailing List Archive (read-only)
1

Dear MNEs,

I just learned about the possibility that the Freesurfer segmentation
quality could depend on the set of imaging parameters used.
For my current measurements I have access to mprages measured on a 3T
Siemens Trio device with a 16-Ch coil.
The segmentation looks ok and also my source estimates, but I was
wondering whether the passing expert settings instead of the defaults
when invoking recon-all should make a practically significant
difference.
Does anyone have experience with this?

Cheers,
Denis

2

Hi Denis,

Well, I'm not sure what you mean by "expert settings". In my
experience, as long as your FreeSurfer white matter surfaces look good
(based on experience with the FreeSurfer Tutorial) and your BEM
surfaces look good (most important for EEG), your results should be
robust. Importantly, at least at the moment the pial surfaces are not
used in MNE, so editing them does not affect the inverse solution.
Obviously optimizing your MR acquisition for the highest quality
possible will improve FreeSurfer's performance (the biggest effect I
have noticed is in reduced manual editing necessary).

The most concerning errors (outside of obvious problems e.g. missing
lobes etc) are those where the surface becomes "sharp". This geometry
will yield strange influences in your estimates and also is
physiologically unlikely. Another important consideration is
decimation (whether you do, and how you do), furthermore, I strongly
recommend using --loosevar with any decimated data.

HTH
D

3

Hi

Thanks for the pointers D,

I was referring to the so called expert preferences section at the wiki pages:
https://surfer.nmr.mgh.harvard.edu/fswiki/recon-all

And I was wondering how save I am with the defaults.

2012/10/10 dgw <dgwakeman at gmail.com>:

Hi Denis,

Well, I'm not sure what you mean by "expert settings". In my
experience, as long as your FreeSurfer white matter surfaces look good
(based on experience with the FreeSurfer Tutorial) and your BEM
surfaces look good (most important for EEG), your results should be
robust.

Ok, good to know that. So that should be fine.

Importantly, at least at the moment the pial surfaces are not

used in MNE, so editing them does not affect the inverse solution.
Obviously optimizing your MR acquisition for the highest quality
possible will improve FreeSurfer's performance (the biggest effect I
have noticed is in reduced manual editing necessary).

The most concerning errors (outside of obvious problems e.g. missing
lobes etc) are those where the surface becomes "sharp". This geometry
will yield strange influences in your estimates and also is
physiologically unlikely.

you mean anatomically implausible junctions or 'spikes'?

Another important consideration is

decimation (whether you do, and how you do), furthermore, I strongly
recommend using --loosevar with any decimated data.

Actually I just sticked with the default parameters exposed in section
3 and 12 of the mne manual which implies --loose 0.2. Actually I
calculated patches on setting up the source space so loosevar could be
an option.
Is it actually possible to exactly tell the differences or make
recommendations for reasonably robust defaults or when to use which
option? For me as an MNE beginner it it still is somewhat difficult to
recognize the consequences of these options.

Thanks,
Denis

4

Hi

Thanks for the pointers D,

I was referring to the so called expert preferences section at the wiki pages:
recon-all - Free Surfer Wiki

Well, I am not a FreeSurfer expert, but from the list I believe they
are mostly for troubleshooting problems (and can be ignored unless you
have a problem that you email the list about and the FreeSurfer team
recommends you use one).

And I was wondering how save I am with the defaults.

2012/10/10 dgw <dgwakeman at gmail.com>:

Hi Denis,

Well, I'm not sure what you mean by "expert settings". In my
experience, as long as your FreeSurfer white matter surfaces look good
(based on experience with the FreeSurfer Tutorial) and your BEM
surfaces look good (most important for EEG), your results should be
robust.

Ok, good to know that. So that should be fine.

Importantly, at least at the moment the pial surfaces are not

used in MNE, so editing them does not affect the inverse solution.
Obviously optimizing your MR acquisition for the highest quality
possible will improve FreeSurfer's performance (the biggest effect I
have noticed is in reduced manual editing necessary).

The most concerning errors (outside of obvious problems e.g. missing
lobes etc) are those where the surface becomes "sharp". This geometry
will yield strange influences in your estimates and also is
physiologically unlikely.

you mean anatomically implausible junctions or 'spikes'?

Well, any error in the estimation should be removed, but some of the
most worrying (important this is my opinion) are the sharp points like
you sometimes see on the optic nerve (
http://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/WhiteMatterEdits
is very helpful). You can often see them most clearly on the inflated
surface. Occasionally it is easier when going through the volume. As
the FreeSurfer team comes out with new versions these problems have
reduced in frequency.

Another important consideration is

decimation (whether you do, and how you do), furthermore, I strongly
recommend using --loosevar with any decimated data.

Actually I just sticked with the default parameters exposed in section
3 and 12 of the mne manual which implies --loose 0.2. Actually I
calculated patches on setting up the source space so loosevar could be
an option.
Is it actually possible to exactly tell the differences or make
recommendations for reasonably robust defaults or when to use which
option? For me as an MNE beginner it it still is somewhat difficult to
recognize the consequences of these options.

I think sticking with that parameter for now is a good idea.