I’m doing coregistration using mne.gui.coregistration().
Locations of the EEG electrodes were aquired using Brainsight software, but unfortunately the locations of the fiducials are not accessible. Therefore, to create the montage I use:
The head model and brain segmentation for MNE analysis were done in freesurfer.
I have tried two different ways to export electrode coordinates from Brainsight:
Export them directly as MNI space coordinates.
Upload individual MRIs to Brainsight, create a head model and fit the electrode positions to this model, thus transitioning to the SUBJECT-SPACE coordinates.
Both approaches give me reasonable fit when running mne.gui.coregistration(), but there are a couple of differences:
In case with MNI space coordinates export from Brainsight, the electrodes are distributed better across the head (e.g., frontline and temporal electrodes fit resembles the actual cap electrode locations better), BUT to electrodes are a bit off the head surface.
In case of SUBJECT-SPACE coordinates from Brainsight, the electrodes are better fit to the skin, but the layout seems a but shrunk (e.g., frontline and temporal electrodes are higher as compared to the real-life cap layout).
Which fit do you find more reasonable?
I personally prefer the first option, since the electrode coverage of the head seems more accurate, but it would be great, if there was an option to project them to the skin surface in MNE environment.
I will be grateful for your feedback!
Please let me know if you require additional details.
unfortunately the locations of the fiducials are not accessible.– Is it for one subject or all? If you have fiducials from other subjects using the same EEG cap, you can average them and add them to your montage info for this subject.
Which montage pattern would reflect a more practical scenario when you put the EEG cap on your subject? Does the electrode coverage fit option 1 or option 2 better? Check the standard layout of the EEG cap for this.
If you see some electrodes are far away from the head even after the ICP fits, you can remove those—in this case, check the omit distance option with (10/5 mm) and check if this reduces the coregistration error values.
Correct me if I am wrong. I see coregistration error shows 7.0 mm mean (..sd) for fig. 1 and 8.0 for fig. 2; this shows figure 1 is a better fit; however, any value of more than 5 mm will likely contribute to poor source localization (if that’s something you are planning to do). Perhaps you can try to optimize the value.
Unfortunately, the fiducials were not saved for the whole sample. I’m currently in contact with the Brainsight reps to find out whether it’s possible to recover them from the digitisation fiels, but it doen’t look like that.
The first montage looks closer to the actual electrodes layout.
In case I omit the electrodes with >10/5mm distance, will they be removed from the subsequent source reconstruction process? If that’s the case, will I have issues with group averaging?
I am aiming to do source reconstruction, therefore adequate coregistration is crucial for me. Do you have any suggestions for the case when I am not able to recover locaitons of the fiducials?