<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
<META NAME="Generator" CONTENT="MS Exchange Server version 6.5.7653.38">
<TITLE>RE: [Eeglablist] EMG artifact question</TITLE>
</HEAD>
<BODY>
<!-- Converted from text/plain format -->
<P><FONT SIZE=2>Thank you very much, Scott. That makes a lot of sense.<BR>
<BR>
Peter<BR>
<BR>
<BR>
-----Original Message-----<BR>
From: eeglablist-bounces@sccn.ucsd.edu on behalf of Scott Makeig<BR>
Sent: Thu 7/30/2009 10:57 PM<BR>
To: Bachman, Peter<BR>
Cc: eeglablist@sccn.ucsd.edu<BR>
Subject: Re: [Eeglablist] EMG artifact question<BR>
<BR>
Peter - You have removed the major eyeblink artifact components (not shown)<BR>
-- Likely they contain no visible high-frequency broadband (EMG-like)<BR>
activity. You are left with the rest of the EEG projecting to the same<BR>
channels, including local scalp muscle signals, which are very likely<BR>
accounted for by a few other ICs. I see no evidence that the (probable)<BR>
muscle activity has increased -- though it has become more clearly visible<BR>
after the large eyeblink potentials have been removed. Of course no spatial<BR>
filter, including those learned by ICA, is ever 'perfect,' but I see no<BR>
evidence here of the increase you are talking about.<BR>
<BR>
Actually, however, the data that remain after removing one or more ICs may<BR>
well actually be larger (e.g., in variance) than the original data. This<BR>
happens when two source with opposite-phase waveforms (in some limited time<BR>
period) sum at a channel, nearly canceling each other out there. In this way<BR>
the summed variance of the individually back-projected ICs is (nearly<BR>
always, with probability approaching 1) larger than the variance of the sum<BR>
of their backprojections, the original data...<BR>
<BR>
Scott Makeig<BR>
<BR>
On Thu, Jul 30, 2009 at 9:35 AM, Bachman, Peter <bachman@psych.ucla.edu>wrote:<BR>
<BR>
> Hi everyone,<BR>
><BR>
> In the course of using ICA in EEGLAB to remove artifacts from EEG data,<BR>
> I've noticed that what appears to be EMG-related contamination will<BR>
> sometimes not only persist in the dataset, but actually appear to become<BR>
> somewhat amplified and even distributed over channels in which it did not<BR>
> appear in the raw data.<BR>
><BR>
> I can imagine why EMG activity may be less likely to collect on to a single<BR>
> component for easy removal - so why it would not be easily removed - but I'm<BR>
> not sure why it would appear to be more of a problem after the raw data is<BR>
> decomposed with ICA and then reconstructed back into EEG channel data (with<BR>
> amplitude scale and other parameters held constant).<BR>
><BR>
> I've attached two .gif images with examples of this phenomenon. In both<BR>
> instances, eye blinks are removed very nicely, but EMG activity seems worse<BR>
> after the cleaning.<BR>
><BR>
> I'm wondering if anyone else has seen this before. If so, do you<BR>
> understand why this might happen and how to avoid it?<BR>
><BR>
> Thank you very much!<BR>
> Peter<BR>
><BR>
> _______________________________________________<BR>
> Eeglablist page: <A HREF="http://sccn.ucsd.edu/eeglab/eeglabmail.html">http://sccn.ucsd.edu/eeglab/eeglabmail.html</A><BR>
> To unsubscribe, send an empty email to<BR>
> eeglablist-unsubscribe@sccn.ucsd.edu<BR>
> For digest mode, send an email with the subject "set digest mime" to<BR>
> eeglablist-request@sccn.ucsd.edu<BR>
><BR>
<BR>
<BR>
<BR>
--<BR>
Scott Makeig, Research Scientist and Director, Swartz Center for<BR>
Computational Neuroscience, Institute for Neural Computation, University of<BR>
California San Diego, La Jolla CA 92093-0961, <A HREF="http://sccn.ucsd.edu/~scott">http://sccn.ucsd.edu/~scott</A><BR>
<BR>
</FONT>
</P>
</BODY>
</HTML>