[Eeglablist] ICA question

[muhammad naeem]

Hi,



Regarding (a) below, would these argument also hold in the case of 
dyadic paradigms (e.g: real-time social/motor coordination)?   I was 
thinking that individual activity modes are likely to be parsed out from
 interacting 
modes thus connecting  joint actions with neuro-electric co-modulation 
and  source localization? Other than the square mixing assumption: 
concatenating two person's data may potentially blow up the sources and 
due to this constraint may force ICA to extract less distinct 
components-  I am wondering how this could be in conflict with 
underlying generative model of ICA?



Best regards,



Naeem.

--- On Fri, 3/2/12, Scott Makeig <smakeig at gmail.com> wrote:

From: Scott Makeig <smakeig at gmail.com>
Subject: Re: [Eeglablist] ICA question
To: "Baris Demiral" <demiral.007 at googlemail.com>
Cc: "eeglablist" <eeglablist at sccn.ucsd.edu>, "Enrico Schulz" <enrico.schulz at gmail.com>
Date: Friday, 3 February, 2012, 2:50 AM

Baris - See >> below. -Scott

On Thu, Feb 2, 2012 at 11:19 AM, Baris Demiral <demiral.007 at googlemail.com> wrote:

Dear Scott,



I should jump in now, and ask you the things that I was curious about.

I cannot wait for hearing your response.



a) What is your position on the evaluation/theoretical feasibility of

the software like EEGIFT etc. (see the e-mail discussion on social

cognition paradigms) and joint ICA. What is your point of view?
 >> Highly sub-optimum. Our result (Delorme, in press PLoS One tomorrow) shows that the more complete the reduction in mutual information between the channels in the component data, the more  components are physiologically distinct, e.g. their maps each represent the projection of a single equivalent current dipole or cortical patch. And ICA across subjects cannot be exact (e.g., cannot fully reduce mutual information).

 

b) How will you guys apply such approaches to ICA clustering etc.?

>> You may also investigate Nima Bigdeley Shamlo's Measure Projection Toolbox for integrating (separately computed!) ICA decomposition measures from a group of subjects and/or sessions. 




c) How will SIFT be affected by this? AND, WHEN will you release new

SIFT package??

>> Tim Mullen is developing SIFT. You may write him directly if he doesn't respond here.



Best,

Baris







On Thu, Feb 2, 2012 at 9:01 AM, Scott Makeig <smakeig at gmail.com> wrote:

> Enrico -

>

> A best solution would be to record from more electrodes at frontal and

> inferior electrodes around the head -- since (muscle) source density is

> higher here, electrode density should best be higher here as well (as

> non-intuitive as that may sound). This is particularly relevant for a gamma

> band study (see Onton & Makeig, 2009 for an example with 256 electrodes over

> the whole scalp).

>

> Also, you should investigate using Amica (Palmer, 2007) with unlikely-data

> rejection set on, and/or possibly using two or more models competing for the

> data (though the latter option is still difficult to interpret). Jason

> Palmer is about to release binaries for PC, Mac, and linux that take

> advantage of multiple cores and processors when possible. A study in press

> (Delorme et al., PLoS One, 2/3/12-) shows Amica to be the best algorithm for

> blind source decomposition of EEG data from at least two angles...

>

> Multi-subject ICA sacrifices much of the specificity and accuracy of ICA,

> and also suffers more from undercompleteness (e.g., more

> distinct-if-overlapping independent source projections than channels) than

> single-subject data....

>

> Scott Makeig

>

> On Wed, Feb 1, 2012 at 6:44 AM, Enrico Schulz <enrico.schulz at gmail.com>

> wrote:

>>

>> Dear EEGlab list,

>>

>> I have a problem with the ICA-based artefact reduction that is actually

>> not just restricted to the EEGlab software.

>>

>> I'm struggling with a lot of high frequency- artefacts at frontal and

>> inferior electrodes around the head exhibiting a much higher amplitude than

>> the cortical gamma band activity I'm interested in. Although it is possible

>> to remove the strongest artefacts, some muscle activity could not be removed

>> in my data sets because some of the artefacts do not give rise to a separate

>> component.

>>

>> In my naive view, in addition to the fact that there are still artefacts

>> in the data set, this could lead to a bias for some subjects. In theory, if

>> a strong artefact gives rise to an independent component and can, hence, be

>> removed, the amount of artefacts in that data set is now lower than in a

>> different data set, where that artefact is not strong enough for a distinct

>> component.

>>

>> The problem is even more complicated if an experimental group (e.g. pain

>> patients) has stronger muscle artefacts than a healthy control group.

>>

>> Sorry for the long introduction, but my actual question is, whether it is

>> possible to concatenate all single subject files and doing the ICA for that

>> big file.

>> I'm aware that this approach has other disadvantages, e.g. it requires a

>> similar topography for each artefact across all subjects and a fast

>> machine.

>>

>> Any help/opinion is highly appreciated!

>>

>> Best regards,

>> Enrico

>>

>>

>>

>> _______________________________________________

>> 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

>

>

>

>

> --

> Scott Makeig, Research Scientist and Director, Swartz Center for

> Computational Neuroscience, Institute for Neural Computation; Prof. of

> Neurosciences (Adj.), University of California San Diego, La Jolla CA

> 92093-0559, http://sccn.ucsd.edu/~scott

>

> _______________________________________________

> 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







--

Ş. Barış Demiral, PhD.

Department of Psychiatry

Washington University

School of Medicine

660 S. Euclid Avenue

Box 8134

Saint Louis, MO 63110

Phone: +1 (314) 747 1603




-- 
Scott Makeig, Research Scientist and Director, Swartz Center for Computational Neuroscience, Institute for Neural Computation; Prof. of Neurosciences (Adj.), University of California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott



-----Inline Attachment Follows-----

_______________________________________________
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/20120203/8b431b53/attachment.html>