I have discovered an electrical artifact that I cannot remove from my experimental setup.
It is fairly regular, however. Thus, It is very easy to remove using ICA when data is collected without a subject. However, ICA on brain data+artifact returns the artifact in many IC’s and so I will have to remove too much brain data with it.
Is it advisable to seed my ICA of brain data with the IC’s created using non-brain data (collected in a way that isolates the artifact itself, basically with electrodes in a bucket of saline)???
Can someone point me in the right direction for implementing this in EEGLAB?
Hi Josh. Interesting thought. I’m not sure if you can feature-guide the ICA decomposition like this (although I don’t follow all of the recent ICA developments). But perhaps a method called joint decorrelation might help – it’s a decomposition method that allows you to design a spatial filter that optimizes some criteria. https://www.ncbi.nlm.nih.gov/pubmed/24990357
We’ve been having a similar problem with a unique recording device. My assumption was that if you have enough data, ICA will isolate a truly independent artifact very effectively. We can’t very well conjure more data, so we discussed making larger epochs and a few other solutions. You can email me back channel and we can include my grad student working on this so you two could discuss your attempts.
Hi Jim. Is that artifact confine to a particular frequency range? If so, you might get a better ICA if you filter only for the artifact.