I'm currently working with the OMEGA dataset and have encountered a recurring issue with some subjects: a few channels in the raw data appear to be particularly noisy — see the example below.
I tried following the Brainstorm-recommended preprocessing pipeline and applied SSP for artifact removal, but it doesn't seem to effectively suppress the noise from these channels. The artifacts are not only visually apparent, but they also severely impact source estimation, especially at specific frequency bands.
I'm wondering:
Should these channels be marked as bad and excluded from further analysis?
Or would it be better to use ICA to try to clean them up instead?
Is there a best practice for handling such channels in the context of group-level MEG analysis?
OMEGA includes many participants from different studies, and the quality of the data has often not been fully validated. There may be however some notes about artefacts in the ..._meg.json file, where there's a SubjectArtefactDescription field (or something similar). For such large artefact, which look like dental work to me, it is very difficult to clean, and depends on how much the head moved as to how well SSP or ICA will perform. You may want to split the data into smaller chunks where the head is still enough, such that the artefact pattern is more consistent - this depends on how much movement there is. You can look at the Brainstorm head motion tutorial for this. But it will likely remain difficult to clean, and it is up to you to decide how much effort you wish to spend on those subjects, or simply to reject them entirely. The artefact is likely present in many sensors, so rejecting so many is not really recommended vs rejecting fewer SSP or ICA components. I'd try these two methods with slightly different settings to find what works best, if you really want to spend more time cleaning them.
P.S. Please avoid including OMEGA participant IDs in public posts. It appears on your figure.
