[Eeglablist] (ask) about averaging and ICA

Bagas Isadewa izadewa at yahoo.com
Thu Dec 30 11:32:09 PST 2010


thank you for the answers, especially for Mr. Andrew Hill and Mr. Tarik S 
I apologize for my ambiguous and unclear question. It caused by my English is 
bad, i'm sorry.

I  will explain about my experiment, now i'm doing my final project to get  
bachelor degree's. I try to generate ERP with subject read an  alphabet/letter 
in a slideshow ( I chosed a/i/u/e/o letter for stimuli).  My goal is to get a 
ERP waveform related to a certain alphabet which  subject has read. Note that 
subject only thinking reading a words not  speak it and not produce any sound. 
The subject just say (or think) the  letter in mind.

for my question number :

1.thank you for  the suggestion, I take just 20 trials for each letter, actually 
I wanted  to do experiment (record many trials) again but i'm  facing a bit of 
problem here. To do an experiment, I must rent an EEG  in medical clinic which I 
must paid some money (unfortunately it quite  expensive). because of that I 
can't do an experiment as many as I  wanted.

2. I mean many N trials in many N epoch (is that right ?) and single-trial in 1 

3.  Let me try to explain what I understand. To extract ERP there are two  ways 
: a. Do the averaging many trials and b. run ICA to the data. And  what I'm 
going to do is run ICA first, then I will average the result of  ICA 

I admit I do not fully understand  ICA, but I still try reading the literature 
(it's little hard in  English, hehehe). There is a question in my mind, how do 
we prove the  result of ICA decomposition is right ? I mean, do the result will 
one  hundred percent represent the true neural signal ? 

thank you very much

bagas isadewa

From: Tarik S Bel-Bahar <tarikbelbahar at gmail.com>
To: Bagas Isadewa <izadewa at yahoo.com>
Cc: EEGLAB <eeglablist at sccn.ucsd.edu>
Sent: Mon, December 27, 2010 8:45:01 AM
Subject: Re: [Eeglablist] (ask) about averaging and ICA

Greetings Bagas,
1. Yes, there are multiple papers that refer to this issue. I would suggest you 
with the HANDBOOKS by Tom Handy or Steve Luck, and these will answer your basic 
questions about the relations
between trial counts, averaged ERPs, and the resultant signal-to-noise ratios. 
You can find these books by doing a Google search, and you can order them from 
Amazon or another bookseller.
Here's the link for the Handy handbook: 
2. Your second question is not clear. It's not clear what you mean by "many 
trials" or only "single-trials" in one epoch.
For information about best practice using EEGLAB and ICA, please read through 
the EEGLAB tutorial online.
A general rule of thumb is that more single trials are better than fewer single 
trials, for an accurate ICA decomposition of the data.
3. Again, you should read and learn basic information about ERP techniques from 
handbooks mentioned in Point 1 above,
or in articles that use ERPs (please search on Google Scholar for examples that 
relate to your research questions). 

It also not clear what you mean by using "only one technique".
You can generate average ERPs without using ICA, and you can also generate ERPs 
after using ICA.
Please be aware that the emphasis in EEGLAB is on single-trial analysis, 
although it can certainly be used for single subject average ERPs.
4. In the future please try to be more specific about what you are trying to do, 
what kind of ERPs you are trying to generate.
Also it seems that you are unaware of the the many potential uses of ICA (for 
artifact detection, for detection of independent modulators, etc..)
Please read through the articles describing EEGLAB (which you can find at the 
EEGLAB site, or via Google Scholar).
Please also read through and do the EEGLAB tutorial, which will answer some of 
your questions.
Then feel free to send some more questions after that.
Good luck Bagas!

On Thu, Dec 23, 2010 at 1:09 AM, Bagas Isadewa <izadewa at yahoo.com> wrote:

>i have several questions, 
>1. if I wanted to average EEG data to yields good ERP, how many minimum trials 
>do I needed ? is there any paper or journals that explained about it ?
>2. then, if I wanted to run ICA to extract ERP, do I need many trials or only 
>single-trials in one epoch will be enough ??
>3. is it necessary to run ICA first then do averaging or it is enough to use 
>only one technique for extracting ERP ?
>if you don't mind please give me an explanation
>thank you very much
>bagas isadewa
>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/20101230/16419e07/attachment.html 

More information about the eeglablist mailing list