I am currently analyzing resting-state MEG data acquired with a CTF system and I had a couple of methodological questions. In several subjects I identified multiple segments of stable head position using Events → Detect head motion (CTF), and I am unsure about the most appropriate way to proceed. After filtering and ICA, one option would be to split the data into the stable-head segments, adjust the head coordinates, and then import them into the database, which would result in multiple runs per subject; however, when computing source time series, this creates issues because the files do not match properly for subject-level averaging. What would be the recommended workflow in this case? Since my main goal is functional connectivity analysis, would it be reasonable to compute FC matrices separately for each stable-head segment and then average them?
Finally, in this scenario, how should bad channels be handled: should they be interpolated, and at which stage of the pipeline?
Why is that? The data would be already in source-level (same anatomy for all the runs).
Both ways can be used, though they respond different questions. Check this other post:
In this scenario, bad channels can be discarded, the source estimation will be done only with the good channels. Interpolated channels do not provide additional information to the model as they are a linear combination of the good channels.