[Eeglablist] Re: ICA Components in eeglab

[Arnaud Delorme]

>
> We had some concerns about the ICA that you could probably help us 
> with.  We are analyzing 37 epochs each about 8.2 seconds long for a 
> 64-ch setup (minus the EMG channels, so about 58 channels).  When we 
> run ICA on the data set, it resolves 58 components.  When we go 
> through all of the components, we do not see the usual eye blink or 
> eye movement components in the first few components or in any of the 
> resulting components.  We thought it might be several things:
>  
> 1)  In previous versions of eeglab for ICA, we were able to specify 
> the number of components.  Even for a 64 channel recording, we usually 
> only asked for 32 components to be resolved.  We speculated that maybe 
> the ICA algorithm is resolving the eye blink/movement component into 
> more than the first few components, making it difficult for us to 
> visually detect.  We thought that if we asked for a lesser number of 
> components, we could see it better.  Is there a way in eeglab to ask 
> for fewer components than the number channels?


To ask for less components, enter the text " 'pca', 32 " in the pop-up 
window for running the ICA algorithm.
I do not know why you did not see the eye blink component. Sometimes 
they do not look standard (see for instance, the tutorial dataset which 
eye blink component has an unusual topography). I think it especially 
occurs with few epochs ?

> 2)  We understand that ICA best runs with is has large data sets to 
> work on.  Could the problem be caused by the limited number of epochs? 

I don't think so. I don't know about your sampling rate but the number 
(37) of epochs is fine. There is a formula on the tutorial page to check 
if you have enough time points.

> Zoltan recalls that when he asked Tzyy-Ping about analyzing large 
> epochs about 4 seconds long, Tzyy-Ping said it had never been tried.

Julie Onton a our lab decomposed some data succesfully with 20-second 
epochs.

> We were considering to cut our Neuroscan CNT files into smaller epoch 
> sizes, ICA them, then put them back together then epoch them again the 
> way we want. 

> However, is there a way to convert the resulting data set back into 
> Neuroscan EEG or CNT formats?  If we can convert them back into ASCII, 
> then Neuroscan's website has downloads to convert them back into EEG 
> or CNT formats. 

It is possible to save your data as ASCII under Matlab. To save the 
current dataset, type

TMP = EEG.data;
save -ascii file.txt TMP

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/20030113/c8239ef3/attachment.htm