Hello Brainstorm community,
I am an EEG researcher using Biosemi 64 channel EEG + Left VEOG and currently deciding which blink artifact removal method (SSP vs ICA) to use. To help me decide I went through tutorials 13 and Artifact cleaning with ICA.
Now, it seems from the tutorials that SSP would work better for MEG and ICA would work better for EEG ocular artifact removal. Upon trying out both methods it seems that the SSP method works better than the ICA method for my data. On that note I have some questions/statements based on the following two comments that I read from the tutorials.
This introduction tutorial will focus on the SSP approach, as it is a lot simpler and faster but still very efficient for removing blinks and heartbeats from MEG recordings.
Based on MY dataset, I think the SSP approach is much simpler and faster but still VERY effective for removing blinks from EEG recordings. I note on several occasions how the SSP approach cleaned all blinks from my data without needing to select additional components to remove. I.e. the first component of each blink "type" was completely sufficient for excellent blink removal.
The SSP technique is not adapted for cleaning events for which we cannot set markers, and in general it is not suitable for low-density EEG.
- In this statement what does "for which we cannot set markers" mean?
- In this phrase "in general it is not suitable for low-density EEG," what level of density are we talking about? 32, 64 or 128 channels?
Finally, what is your recommendation for citing individual tutorials?
Many Thanks.
MinChul Park
ICA works both for MEG and EEG. But for MEG it can be extremely time consuming with 300 sensors, and then artifacts are then spread over tens of components and it gets difficult to identify them.
SSP on EEG can remove too much signal because the topographies are a lot smoother. The first component typically captures a lot more than the targeted artifact.
Based on MY dataset, I think the SSP approach is much simpler and faster but still VERY effective for removing blinks from EEG recordings.
Good if you're happy with it. But be careful with the amount of extra signal you are removing, especially with only 64 channels. You may end up removing a large share of other brain-generated components as well (e.g. the alpha generators).
It's often worse to remove too much signal than to keep the artifacts.
I agree with what Arnaud Delorme recently published: EEG is better left alone.
https://www.nature.com/articles/s41598-023-27528-0
In this statement what does "for which we cannot set markers" mean?
= For which there is not an event marker that indicates each occurrence of the artifact.
= either continuous noise, or artifacts that are too complicated to detect from the recordings.
In this phrase "in general it is not suitable for low-density EEG," what level of density are we talking about? 32, 64 or 128 channels?
32 would be considered low-density, 128 and 256 high-density.
64: not sure.
This all depends on the recordings quality, acquisition filter, amount of noise, etc. It has to be evaluated on a case by case basis.
Thank you for your reply Francois.
I agree with your comments about being careful with the amount of extra signal that I could be removing. And thanks for the article - really useful stuff.
On that matter, how do I cite the Brainstorm tutorials? Any suggestions?
As always thank you so much for all your work in answering the hundreds of questions you might be getting every day as well as developing such a useful tool to analyse MEG/EEG effectively and efficiently. I really appreciate your work.
Cheers.
Thank you for these kind words.
how do I cite the Brainstorm tutorials?
Use the URL of the page you are referring to.
And cite this article in your publications: Brainstorm: a user-friendly application for MEG/EEG analysis - PubMed
Thanks