[FieldTrip] ICA+frqanalysis questions

odelia nakar odidodi at hotmail.com
Wed Jun 1 09:10:30 CEST 2011

Hi all,

Thank you all for the discussion. I'm sorry I response only now, I went on a vacation...
I actually didn't want to run the ICA on trials with blinks, rather on all trials, but to remove the component only for trials with a blink. Your critics concern my idea as well, since if I have any kind of correlation between condition and blink I might yield a difference that does not exist between conditions...
Considering your suggestion, Mahesh, unfortunately I don't have EOG recording for this experiment.

Thanks again for the discussion,

Date: Fri, 27 May 2011 00:32:01 -0400
From: mahesh.casiraghi at gmail.com
To: yuvharpaz at gmail.com; fieldtrip at donders.ru.nl
Subject: Re: [FieldTrip] ICA+frqanalysis questions

Dear Yuval and discussion group,

it seems to me that what you are proposing is getting close to what proposed by the hybrid approach of regica described here: 

Manousos A. Klados, Christos Papadelis, Christoph Braun, Panagiotis D. Bamidis, REG-ICA: A hybrid methodology combining Blind Source Separation and regression techniques for the rejection of ocular artifacts, Biomedical Signal Processing and Control, In Press, Corrected Proof, Available online 16 March 2011, ISSN 1746-8094, DOI: 10.1016/j.bspc.2011.02.001.

They suggest to selectively run regression based AR only on those components which correlate with EOG signals. This makes sense to me and I have been trying to experiment that on some old data, although with no clear conclusions yet. It may be worth a try for Odelia: Anybody out there with some insights for this - or maybe a similar - approach? 



Mahesh M. CasiraghiPhD candidate - Cognitive Sciences
Roberto Dell'Acqua Lab, University of PadovaPierre Jolicoeur Lab, Univesité de Montréal
mahesh.casiraghi at umontreal.ca

I have the conviction that when Physiology will be far enough advanced, the poet, the philosopher, and the physiologist will all understand each other.
Claude Bernard

On Thu, May 26, 2011 at 11:40 PM, Yuval Harpaz <yuvharpaz at gmail.com> wrote:

Dear discussion groupDid anybody consider smoothing or filtering the component trace before rejecting it?
it seems that the added noise to no-blink trials is in a frequency higher than that typical to blinks. what if we evaluate the component weight, creating a trace for the eyeblink component for every trial, then bandpass filter the blink trace , say 0.1-25Hz, and only then remove the component from the data?

On 27 May 2011 06:16, Joseph Dien <jdien07 at mac.com> wrote:

Stefan, just to be clear, I don't think any of us were saying not to use ICA to correct blinks.  David was just saying that there are potential concerns when one only applies the ICA to the blink trials rather than to all the trials.  I myself use EEGlab's infomax implementation in the automatic eyeblink correction tool of my EP Toolkit (http://sourceforge.net/projects/erppcatoolkit/).

Now that said, I should add a little more nuance to my response.  One of the things I observed (or rather, that Tim Curran pointed out to me) is that when you apply ICA to remove eyeblink artifacts in this manner, it can actually substantially increase the noise level in the data, so for the trials without eyeblinks it can have a considerable cost.  So in order to balance the cost/benefit ratio, what I did was to include a trial by trial criterion that the putative eyeblink factors would only be removed if doing so reduced the overall variance of the trial.  This approach does still have some potential for causing the concerns that David raises but not as much as only applying the ICA to blink trials since it does end up getting applied to non-blink trials too.  This does mean that one should be cautious about any apparent effects in the artifact corrected data that are centered around the eyes (that have a blink topography) but that goes without saying in any case.  So anyway, I agree, it's not perfect but seems to be the best available option.

On May 24, 2011, at 3:11 AM, Stefan Debener wrote:

    Hi Odelia,


    I have a slightly different opinion here. It is certainly true that
    any filter has the tendency to distort data (with distortion I mean
    that data consists of a mixture of some wanted, true signal and some
    unwanted signal, and that the removal of the unwanted part of the
    signal is neither complete nor specific). In our lab we regularly
    use ICA for artefact removal (and more), and the benefit/gain is
    clearly are much larger than the distortion. In fact there are a
    number of examples out showing that currently only ICA (or related
    tools) can recover the study of (a substantial fraction of the
    wanted) EEG signal (but again, it is NOT a perfect tool at all), in
    particular in cases where other means of SNR enhancement don't work
    well (averaging, spectral analysis). I am happy to provide
    references if you are interested...


    For the evaluation of outcome it would be reasonable to not evaluate
    the ERP alone, as this could be misleading. Better evaluate the
    sensitivity and specificity of an eye blink attentuation approach on
    the single trial (and single subject) level, this will give you good
    insight. And it is worth keeping in mind that the preprocessing of
    the data (among other issues, like the quality of the recording and
    so on) largely determines the quality of the output (for some
    introduction you may look up chapter 3.1 in Ullsperger &
    Debener, 2010, Simultaneous EEG and FMRI, Oxford University Press).
    Just by a different preprocessing ICA output could vary between crap
    and excellent unmixing. Thus a poor ICA eye blink attenuation would
    make me a bit suspicious...






    Am 5/24/11 4:00 AM, schrieb Alexander J. Shackman:
      And for a related perspective, see 


      McMenamin, B.
        W., Shackman, A. J., Greischar,
        L. L. & Davidson, R. J. (2011). Electromyogenic
        artifacts and electroencephalographic inferences revisited, Neuroimage, 54, 4-9. 





      On Mon, May 23, 2011 at 8:07 PM, Joseph
        Dien <jdien07 at mac.com>

            I agree with David's reasoning.  You may find the
              following article to be of help as well in understanding
              the issues involved:

              Dien, J., Khoe, W., &
                Mangun, G. R. (2007). Evaluation of PCA and ICA of
                simulated ERPs: Promax versus Infomax rotations. Human
                  Brain Mapping, 28(8), 742-763.




              On May 23, 2011, at 11:57 AM, David Groppe wrote:

                Hi Odelia,

                    When you use ICA (or any other spatial filter) to
                  correct for EEG

                  artifacts, you're going to distort your data some by
                  removing true EEG

                  activity in addition to the artifact (for an
                  explanation, see:


                  So to minimize distortion, it would be better not to
                  apply ICA

                  artifact correction to artifact-free data.  However,
                  if the frequency

                  of the artifact differs across experimental
                  conditions, it could

                  confound your analysis.  For example, I suspect people
                  blink more

                  often to targets in an oddball experiment than
                  standards.  Thus if you

                  apply ICA only to blinky trials, you could find a
                  difference between

                  the EEG response to standards and targets that simply
                  reflects the

                  fact ICA removed more EEG activity in the target
                  trials (i.e., it

                  wouldn't reflect a true difference in neural

                       hope this helps,



                  On Mon, May 23, 2011 at 1:44 AM, odelia nakar <odidodi at hotmail.com>

                  Hi all,


                  I'm troubled by the fact that
                    when I use ICA for blinks\eyes movements

                    removal, I remove the relevant components also from
                    trials that do not

                  contain blinks\eyes movements.
                    In order to avoid this bias we thought to

                    combine the data before ICA ("data" structure) with
                    the data after ICA

                  ("dataica" structure), only in
                    specific trials, as follows:





                    blinks=[2 4 5 8 bla bla 156];

                  for ind=1:length(blinks)




                  To my first question: I just
                    wanted to check that there is no problem with

                  that, or any reason not to use


                  Another issue- I use motor
                    learning task, and I'm trying to understand what

                  happens through the process,
                    in terms of power-frequency changes through the

                  process. How would you
                    recommend that I'd use the ft_freqanalysis function?

                  What method to use (or what do
                    I need to consider when choosing the method




                  Thanks a lot,




                  fieldtrip mailing list

                    fieldtrip at donders.ru.nl







                  David Groppe, Ph.D.

                  Postdoctoral Researcher


                  Dept. of Cognitive Science

                  University of California, San Diego




                  fieldtrip mailing list

                  fieldtrip at donders.ru.nl





                                                    Joseph Dien
                                                    E-mail: jdien07 at mac.com
                                                    Phone: 301-226-8848
                                                    Fax: 301-226-8811










          fieldtrip mailing list

          fieldtrip at donders.ru.nl






      Alexander J. Shackman, Ph.D.

      Wisconsin Psychiatric Institute & Clinics and

      Department of Psychology

      University of Wisconsin-Madison

      1202 West Johnson Street

      Madison, Wisconsin 53706


      Telephone: +1 (608) 358-5025

      Fax: +1 (608) 265-2875

      Email: shackman at wisc.edu


fieldtrip mailing list
fieldtrip at donders.ru.nl


Prof. Dr. Stefan Debener
Neuropsychology	Lab
Department of Psychology
University of Oldenburg
D-26111 Oldenburg

Office: A7 0-038
Phone: +49-441-798-4271
Fax:   +49-441-798-5522
Email: stefan.debener at uni-oldenburg.de


fieldtrip mailing list
fieldtrip at donders.ru.nl


fieldtrip mailing list

fieldtrip at donders.ru.nl




a link to the BIU MEG lab:
"Many were increasingly of the opinion that they'd all made a big 
mistake in coming down from the trees in the first place. And some said 
that even the trees had been a bad move, and that no one should ever 
have left the oceans". Douglas Adams


fieldtrip mailing list

fieldtrip at donders.ru.nl


fieldtrip mailing list
fieldtrip at donders.ru.nl
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110601/ec53b370/attachment-0001.html>

More information about the fieldtrip mailing list