[Eeglablist] Any tips for unsupervised (sic) ICA ECG / EOG component, rejection?

[Jan R Wessel]

Dear Panagiotis,

have you tried Felipa Viola's EEGlab plugin CORRMAP? (Viola et al., Clin 
Neurophys 2009)
I use it for horizontal and vertical eyemovements all the time and it 
works like a charm (with a few minor tweaks, like removing the 3 IC 
maximum constraint when using non-PCA preprocessed high-density arrays 
which sometimes results in overfitting, i.e. the representation of 
eyeblinks in more than 3 ICs).

If I remember correctly, they also describe that it works quite well for 
ECG artifacts in their paper. So I highly recommend you check it out 
(both the paper and the algorithm. It's mainly based on correlations, 
but has a neat way of determining a cutoff threshold).
You can call CORRMAP from the EEGLAB GUI under the "study" option, I think.
Type "edit corrmap" into your command line to access the function itself.

Also, there have been a couple more papers on the automated rejection of 
artifacts based on IC properties, one from Robert Whelan / H. Nolan and 
colleagues of Dublin, called FASTER (J Neurosci Meth 2010), and one 
called ADJUST (Mognon et al., Psychophys 2010).
I haven't used the latter two methods yet, but those papers are also a 
good place to look for ideas for automated rejection (I think FASTER 
uses a couple of z-thresholded metrices).

I also remember two people from SCCN trying to establish a pattern 
recognition plugin for this exact purpose, but I don't know whether this 
is out yet. But I can assure you, once it is, it'll probably be amazing.

Anyhow, have fun with automated IC rejection / selection, it's a very 
interesting topic indeed.

Cheers,
Jan

-- 

Dr. Jan R. Wessel
Cognitive Neurology Research Group
Max Planck Institute for Neurological Research
Gleueler Strasse 50, 50931 Koeln
http://www.nf.mpg.de/cv/jan-wessel.html