[Eeglablist] bad channels removal before ICA

[Arnaud Delorme]

>
>
>1, I do ICA on averaged data. The averaging program can detact and
>remove bad channels. I have two questions here. 1.1, after averaging,
>the bad channels are replaces by 0 if they are bad most of the time.
>What should I put in ICA? I don't think I should put 0 in it. So should
>I use the averaged data from adjacent channels? 1.2, The averaging
>program seems to be able to detact and remove eye blinks data also. Then
>what the meaning of doing an ICA on the data?
>
You should neither use 0 nor interpolate the activity for bad channels. 
You should simply remove them by visual inspection.

>2. I do ICA on single trials. Should I just remove the bad channels by
>visual examination? If so, the same question as above
>
Then, the same response as above ;-)

>Finally, I actually don't know whether I need to do anything on the bad
>channels before ICA. I guessed ICA could separate the bad channels out
>by different components. But it is not proved by my work. When I put
>data including some bad channels in ICA, I tried to get 128 components.
>At least the first 10 components (having the largest variance) only
>represent the bad channels. I don't think it's right to remove the 10
>largest components in ICA, is it?
>
You are correct. You should not remove these components but rather remove 
bad channels prior to running ICA (because your ICA decomposition might be 
strongly biased by these bad channels).


>By the way, should ICA be applied on averaged data or single trial? If
>on single trial, I still need to average it in the later ANOVA. So
>what's the difference?
>
Some earlier studies applied ICA to collections of single-trial EEG 
data averages, but this raises several problems. First, ICA may require 
many observations to separate two or more processes, so a problem 
often faced using averaged EEG data is that there are not enough 
conditions in the training set to obtain stable ICA components. 
Another problem with using averaged EEG data is that the averaging process
may cancel out the activity of many brain sources. Finally data averages 
by their nature contain sums of activity occurring at similar times. 
When two or more sources reliably contribute to the response average at 
the same times, ICA may assign their sum to a single component.
[From Delorme et al, 2001; http://www.sccn.ucsd.edu/~arno/mypapers/DelormeCNS2001.PDF]

Best 

Arno


-- 

*Arnaud Delorme, Ph.D.*
Computational Neurobiology Lab, Salk Institute
10010 North Torrey Pines Road
La Jolla, CA 92037 USA

*Tel* : /(+1)-858-458-1927 ext 15/
*Fax* : /(+1)-858-458-1847/
*Web page *: www.sccn.ucsd.edu/~arno <http://www.sccn.ucsd.edu/%7Earno>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20031023/c4bfbd7c/attachment.htm