I am using ICA to repair artifacts in a sample dataset. When plotting ICA component properties, how should I interpret the epoch variance plot and histogram? How should this inform my identification of ‘bad’ independent components? Any guidance would be greatly appreciated.
I am new to EEG data so I apologise if this is a basic question.
Hello, I believe @sappelhoff, @cbrnr, and @sophie are regular users of ICA, maybe they can give you some good advice here!
To be honest, I don’t really looks at these plots that much. I find the scalp topographies and PSD much more informative (and sufficient to identify e.g. ocular components). However, you can use the segments variance plot to identify segments with high variances; these likely contain large artifacts, which you should remove from the original raw data. A “normal” plot would show a normal distribution of variances I believe.
I also don’t look at these too much. I usually correlate my VEOG, HEOG and ECG channels with the IC timecourses to get an “automatic first guess” of which of these are the corresponding bad components, and then I double check by inspecting the IC topographies and IC timecourses visually.
A big help to learn more on how to label components was this resource:
And of course the mne tutorial:
I’d like to add that @adam2392 and others have just recently published MNE-ICALabel so ease automated labeling of ICA components.
Thanks for your help I will check this out!
I’ve added these variance plots originally mirroring the plots and suggestions in:
Hipp, J. F., & Siegel, M. (2013). Dissociating neuronal gamma-band activity from cranial and ocular muscle activity in EEG. Frontiers in human neuroscience , 338.
Hipp and Siegel suggested these plots are useful when looking for “nice” muscle artifacts, and I agree they sometimes help: