[Eeglablist] Massimiliano Papera removing ICA component

[Tarik S Bel-Bahar]

in addition to my response further below, here are
some other eeglab discussions I found via google (searcing PCA, ICA, eeglab)
that seem related to  some of your questions, and that you may find useful:

[*Eeglablist*] Problems with rank computation -
*ICA*<http://www.google.com/url?sa=t&rct=j&q=site%3Asccn.ucsd.edu%20running%20ica%20with%20pca%20eeglab&source=web&cd=14&ved=0CC8QFjADOAo&url=http%3A%2F%2Fsccn.ucsd.edu%2Fpipermail%2Feeglablist%2F2012%2F004652.html&ei=5Vh3UOXCDMbiiAKIzIDoCA&usg=AFQjCNFrJ_1rUMANWhgYkho9LdIAFfYTmg&sig2=b6FCefIns3Du6J3o8CxbTA>
sccn.ucsd.edu/pipermail/*eeglablist*/2012/004652.html

[*Eeglablist*] value of *PCA* pre-processing before *running ICA* on EEG *
...* <http://sccn.ucsd.edu/pipermail/eeglablist/2006/001475.html>
sccn.ucsd.edu/pipermail/*eeglablist*/2006/001475.html

[*Eeglablist*] The number of the *ICA* components is smaller than the
*...*<http://sccn.ucsd.edu/pipermail/eeglablist/2011/003768.html>
sccn.ucsd.edu/pipermail/*eeglablist*/2011/003768.html

[*Eeglablist*] A second *ICA* after removing
components?<http://sccn.ucsd.edu/pipermail/eeglablist/2011/004269.html>
sccn.ucsd.edu/pipermail/*eeglablist*/2011/004269.html

[*Eeglablist*] Computing a second *ICA* on a dataset (Conny
Kranczioch)<http://sccn.ucsd.edu/pipermail/eeglablist/2008/002541.html>
sccn.ucsd.edu/pipermail/*eeglablist*/2008/002541.html

and I assume you have already read through the tutorials, worked with the
sample data provided, have done some exploring on your own, and that you
have read at least
10 articles that use the methods you are using. See Google Scholar for help.
See also this recent article from the EEGLAB group:

 PLOS ONE: Independent *EEG Sources* Are
*Dipolar*<http://www.google.com/url?sa=t&rct=j&q=eeg%20source%20dipolar%20makeig&source=web&cd=1&ved=0CCQQFjAA&url=http%3A%2F%2Fwww.plosone.org%2Farticle%2Finfo%253Adoi%252F10.1371%252Fjournal.pone.0030135&ei=el13UKycGILOiwLi_IDQAw&usg=AFQjCNEe9tgWzzASZLrsCdwcHCN5Guc6aQ&sig2=qwyYy38Vv57pW11aXBvxjw>
www.plosone.org/.../info%3Adoi%2F10.1371%2Fjournal.pone.0030...



On Thu, Oct 11, 2012 at 4:44 PM, Tarik S Bel-Bahar
<tarikbelbahar at gmail.com>wrote:

> Greetings M,
>
> ***we could use an opinion from an ICA/PCA expert here, but here are some
> thoughts. *******
> I have tried to keep it as clear, simple, and (hopefully) useful as
> possible.
>
> I assume you have done some cleaning before running your first ICA,
> as per general pre-ICA guidelines.
> If so, then your sequence of
> first ICA, then reject some epochs using IC data,
> then second ICA, then remove some bad components...
> seems OK. This seems like a normal processing procedure.
>
> So now you can analyze specific ICs,
> or you can "reconstruct" your eeg data using only the remaining ICs,
> and have "cleaner" EEG data for "normal" EEG analyses.
>
> imho, I don't see why you want to re-run the ICA again now for the third
> time.
> If you do, you might get a slightly better decomposition,
> but there should not be a significant change of ICs.
>
> on your question about PCA, perhaps one of the experts can make a
> suggestion,
> but from my understanding, the PCA option is used when
> A. the "dimensionality" and/or the "rank" of the data has been reduced via
> some method,
> and you want ICA to know this before running a new decomposiiton
> B. or you want to "reduce" the total number of dimensions that ICA
> generates,
> for some other good reason. [see published articles on this point]
>
> Last, I want you to know I am not 100% sure about whether or not you "have
> to" run ICA with PCA options
> when you are re-running ICA after having "dropped/removed" some
> components.
> one previous eeglab list discussion:
> http://sccn.ucsd.edu/pipermail/eeglablist/2012/004543.html
>
> Good luck and let the list know of your success.
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20121011/b9ef26fb/attachment.html>