[Eeglablist] running ICA a second time

[Veerle ROSS]

Thank you! This is very helpfull Regards Veerle



Arnaud Delorme <arno at ucsd.edu> schreef: 


Dear Veerle,

I was just reviewing a PhD thesis (http://infosci.otago.ac.nz/carl-leichter-phd) that showed using simulations and real data that removing PCA components creates artifacts in the data, especially in the power spectrum.


This thesis shows that it is not OK to remove "noise" using PCA when processing data channels. So it is even worse when using a statistical algorithm like ICA. Removing PCA components alters the linear projection of EEG sources to scalp sensors that ICA sought to extract. 


Arno

On 25 Feb 2013, at 01:16, Veerle ROSS wrote:

Dear Arno
Thank you for the advice. May I ask why it’s a problem to run ICA with PCA? There are some references available that recommend to use PCA before ICA.
See for instance;
Leonid Zhukov, David Weinstein and Chris Johnson (2000): Independent Component Analysis For EEG Source Localization In Realistic Head Models.
è The EEG data is first decomposed into signal and noise subspaces using a Principal Component Analysis (PCA) decomposition. This partitioning allows us to easily discard the noise subspace, which has two primary benefits: the remaining signal is less noisy, and it has lower dimensionality. After PCA, we apply Independent Component Analysis (ICA) on the signal subspace.
That’s why we thought that we could use ICA the first time to remove the noise and a second time to remove the remaining artifacts.
Best Veerle
 
From: Arnaud Delorme [mailto:arno at ucsd.edu] 
Sent: maandag 25 februari 2013 2:19
To: Veerle ROSS
Cc: eeglablist at sccn.ucsd.edu
Subject: Re: [Eeglablist] running ICA a second time


 
Dear Veerle,
 

you can run ICA twice in a row and use the first ICA to detect bad portion of data or bad channels (because it is often easier to detect these when looking at ICA components). After you have removed bad portion of data or bad channels, you may rerun ICA.

 

But you should not remove ICA components, then rerun ICA (the data dimension would be reduced and this will force ICA to run PCA first - assuming it is able to automatically detect the dimension reduction). In other words, do not do it.

 

Thanks,

 

Arno

 
On 19 Feb 2013, at 01:36, Veerle ROSS wrote:




Hi all

For a current study we are applying ICA. Because our lab was not free of noise we wanted to run ICA 2 times and we hope to obtain cleaner components the second time.

After searching the internet and the literature we are still not sure what the limits of such a second ICA run are.

Therefore we were wondering:

1.       Is it even possible to indicate ICA components in both the first and second ICA

a.       We read that you can use the PCA option for this to reduce the dimensions

b.      However why can’t we use ICA the second time?

c.       Other sites mention that you shouldn’t indicate components in the first ICA

2.       If we run ICA 2 times

a.       Should we only remove the noise the first time?

b.      Or should we also already indicated components such as blinks, bad channels, etc.

Thanks in advance.

Regards, Veerle Ross

 

_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu


 






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20130225/f70dee8a/attachment.html>