[Eeglablist] Conducting ICA on correct or all epochs

Scott Makeig smakeig at gmail.com
Sat Dec 4 17:42:36 PST 2010


Baris -

Below did you leave out a step '6.5 Back-project the remaining ICs to the
scalp channel space, then select a channel of interest.' ?

If so, you may be ignoring the strongest part of ICA-based filtering, namely
looking for significant effects at the single IC source level. Of course
here one wants then to compare ICs (and their effects) across subjects, for
which a great deal of machinery exists in EEGLAB.

Best,

Scott Makeig


On Wed, Dec 1, 2010 at 1:23 PM, Baris Demiral <demiral.007 at googlemail.com>wrote:

> Dear David,
>
> Thank you for your e-mail. I should mention that I am enthusiastically
> following your work on the reliability of the ICA and other EEG related
> issues. I also plan to use Mass Univariate ERP toolbox sometime in future.
>
> It took me a while to test and play with the data. Well, here is what I
> have done, and my evaluation:
>
> I had max 40 trials from each condition. Participants made mean of 6
> incorrect decisions per condition, so the odds of answering a question
> correctly was 34/6. I analyzed the data in this order:
>
> 1-Filtered, re-referenced (bi-mastoid), epoched data goes into ICA (epochs
> with very gross artifacts are removed as suggested by ADJUST algorithm
> leading to around 1 or at most 2 epochs to be rejected)
> 2- Baseline to -200-0ms
> 3-If correct trials should be used, then select the correct epochs and go
> to 5, else go to 4
> 4-Use all trials
> 5- Use ADJUST algorithm to detect and remove problematic ICs automatically
> 6-Take out the incorrect trials if there are any
> 7-Export data for statistics
>
> I also ran the classical method of rejecting the epochs with over 100
> microvolts observed on the EOG electrodes (H1, H2 and VA2) before
> baselining, leading to the elimination of mean of 6-7 epochs per participant
> most probably due to the eye blinks. But, note that even though this method
> is quiet common, it somehow ignores the individual EOG potential difference
> strengths (some subjects might have eye blinks less than 100 microvolts
> sometimes).
>
> When I used all the epochs for feeding ICA, the statistical output (ran
> ANOVA) was almost mirroring the statistical output I obtained from the
> classical method, and the components which I was suspecting of being
> artifacts disappeared. When I used the correct trials only, some earlier and
> later components (mainly centro-temporal components which I believe not
> artifacts) disappear/attenuated and in one case one new components appeared
> mainly frontally.
>
> My overall experience suggest that using 40 epochs per condition including
> the incorrect epochs might be better IF you are only concerned about
> 'artifact correction' via a toolbox like ADJUST which is mainly depending on
> the ICAs. Ignoring ICA reliability issues and assuming ADJUST treats the
> data similarly every time, I think we need to use more trials.
>
> best,
> Baris
>
> On Thu, Nov 11, 2010 at 3:33 PM, David Groppe <dgroppe at cogsci.ucsd.edu>wrote:
>
>> Hi Baris,
>>   ICA's performance will generally degrade as the number of
>> electrical sources increases (see
>> http://www.cogsci.ucsd.edu/~dgroppe/PUBLICATIONS/GroppeCSO2008.pdf).
>> The incorrect trials probably have some EEG activity not present (or
>> at least less present) in the correct trials.  So if you have
>> sufficient data to run ICA on just the correct trials, it would
>> probably be better just to use the correct trials.  If you don't have
>> enough data using just the correct trials though, you'll probably be
>> fine using the all the trials, since surely a lot of the EEG activity
>> is common to both sets of trials.
>>         hope this helps,
>>             -David Groppe
>>
>>
>> On Wed, Nov 10, 2010 at 4:19 PM, Baris Demiral
>> <demiral.007 at googlemail.com> wrote:
>> > Hi everyone,
>> > I am running a simple EEG experiment where I measure reaction times and
>> > accuracies.
>> > I want to use ICs for artifact removal, and I will report only the
>> correct
>> > trials (hits).
>> > So would it be better to use the correct epochs for the ICA to correct
>> for
>> > the artifacts or is it OK to use all the epochs to detect the artifacts
>> and
>> > then run the artifact correction (pop_subcomp) followed by deleting the
>> > incorrect epochs?
>> > Thanks,
>> > Baris
>> >
>> > --
>> > SB Demiral, PhD.
>> > Department of Psychology
>> > 7 George Square
>> > The University of Edinburgh
>> > Edinburgh, EH8 9JZ
>> > UK
>> > Phone: +44 (0131) 6503063
>> >
>> > _______________________________________________
>> > 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
>> >
>>
>>
>>
>> --
>> David Groppe, Ph.D.
>> dgroppe at cogsci.ucsd.edu
>> http://www.cogsci.ucsd.edu/~dgroppe/
>>
>
>
>
> --
> SB Demiral, PhD.
> Department of Psychology
> 7 George Square
> The University of Edinburgh
> Edinburgh, EH8 9JZ
> UK
> Phone: +44 (0131) 6503063
>
> _______________________________________________
> 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, University of
California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20101204/bcb36a6d/attachment.html 


More information about the eeglablist mailing list