Coherence on averaged electrode space

Dear Francois,

sorry for bothering but I have a question regarding coherence.

I was planning on commuting coherence on my resting-state EEG data however, I have 256 electrodes and I was wondering whether I can do it by reducing the electrode space (so that I have for example group of FCz electrodes, group of Fp1 electrodes etc.). Is it possible to do it via Brainstorm? What would be the pipeline for that? It's my very early attempt to dive into connectivity, thus I'm lacking experience.

I would be very grateful for any help! And thank you!

Unfortunately, we do not have (yet) recommendations to share regarding the computation of sensor-level EEG connectivity by groups of electrodes.

If you want to group your electrodes (averaging them?) you can do so by creating a custom montage, and then running the process "Apply montage" on the imported recordings.

Coherence is not so long to compute, maybe keeping the full 256x256 coherence matrices and doing some manual post-processing later could be a better approach.

thank you very much! My problem does not lie in computing - it can be done fast. However, interpretation gives me a headache... Thus I thought that making "electrode ROIs" would make it cleaner and more interpretable :slight_smile: I hope I'm not making a conceptional mistake here :slight_smile:

Thank you again!

It makes sense, but if you average EEG signals before computing the connectivity measure, you might lose some of the fine frequency content of the signals.
And if you want to do your ROI analysis after estimating the 256x256 coherence, then you need to figure out a way for summarizing all the pairs of interactions within to clusters of electrodes. Averaging would work, but I don't know if this approach has ever be published anywhere.

@Sylvain @hossein27en @Raymundo.Cassani
Any recommendation to share?

Exactly! - how to sum it up after extracting 256x256 coherence matrix? I'm struggling to come up with the idea - so ROIs before coherence were my first thought... But again I agree that I'm loosing quite a bit of info that way...

If anyone has any idea - please share! :slightly_smiling_face::slightly_smiling_face:

Hi again,

I think I have found a solution, so I'm sharing it here. It seems that computing data-driven functional units can do the trick. Here is a paper which shows how to do it:

Then, there is an interesting option of comparing the FU maps using graph characteristics:

Personally, I tried to implement it but - as it happens with learning new approaches - bug comes after bug :wink: I hope that eventually it works :slight_smile:

@Francois thanks again for your involvement!