External Email - Use Caution
Dear FreeSurfer team,
I'm writing to report a coordinate-frame issue when using recon-all-clinical (FreeSurfer 8.1) with downstream MEG/EEG coregistration and BEM workflows (MNE-Python), along with findings from testing across multiple subjects.
I have a MEG processing pipeline that has worked reliably using standard recon-all (FreeSurfer 7.3). After switching to recon-all-clinical (FS 8.1) on the same data, the EEG/MEG sensors appear completely displaced from the scalp surface during coregistration.
I tested the watershed BEM generation across 8 subjects using recon-all-clinical output and found the following: - Using norm.mgz (the SynthSR-derived volume from recon-all-clinical) for watershed BEM consistently produces surfaces with inverted normals. The norm.mgz has a non-standard geometry compared to what the standard pipeline produces. - Using a separately conformed T1 (via mri_convert --conform on the original input) for watershed BEM produces correct topology and passes all validation checks. - The cortical surfaces from recon-all-clinical live in norm.mgz space. However, when using the conformed T1 watershed for BEM, 7 out of 8 subjects showed 99-100% source retention, indicating the coordinate frames are close enough for practical use. - Simply symlinking T1.mgz to norm.mgz fixes the coordinate alignment but introduces the inverted normals problem, so this is not a viable workaround.
1. Is this a known issue? 2. Is there a more principled approach than the workaround above for reconciling the coordinate frames between the clinical pipeline surfaces and the conformed BEM surfaces?
Thanks! Noam Peled