[Eeglablist] ICA for eye movement rejection
Tarik S Bel-Bahar
tarikbelbahar at gmail.com
Sat Mar 17 17:27:41 PDT 2012
Greetings, Some quick thoughts, may they be of use:
it seems you are saying that although you get ICs that are mainly blink
activity or lateral eog activity,
your remaining components still contain what looks like eye-blink activity,
suggesting the decomposition is not doing that well.
For best ica results, you of course need enough timepoints to meet ICA's
and you need to make sure that the worst artifactual periods (not the
are removed before doing your first ICA.
Also, examine what different results you get during your post-first-ICA
when you use ICA-based artifactual epoch rejection (it was not clear
whether you are
artifact-detecting with eeg or components at that second step).
I assume you are also removing bad channels (not frontal eye channels
unless they are very noisy).
I would also suggest using a less intense threshold that the +-75 one you
If this is dirty pediatric data, let's treat it as gingerly as possible.
Another possible issue is total number of trials that you have.
If too low a number of trials are available to begin with, this
can cause a problem with too many bad trials,
or with not enough data for good ICA.
On Thu, Mar 15, 2012 at 9:30 AM, Budd, Mary-Jane <mbudd at essex.ac.uk> wrote:
> Dear All,
> I know that this has been a much discussed topic but I am rather confused
> some of the responses. I am looking at children's datasets that include
> eye movements (both blinks and horizontal movements). I have run ICA and
> identified eye blinks and distinct muscle artifacts. If I remove these
> components and then run an automatic rejection procedure (+-75microvolts on
> all electrodes) over half the epochs are rejected due to there still being
> eye movements on the eye and frontal electrode channels. I have read that I
> should not remove the components but instead scan the components and remove
> 'noisy' epochs (I assume this means removing epochs where eye blinks are
> present). RE-running the ICA will then result in 'cleaner' components which
> hopefully will remove the eye movements from my data. I have a couple of
> questions regarding this:
> 1. It would be good to avoid removing epochs as the children blink a lot
> and so I am likely to lose much data.
> 2. I thought this was the benefit of using ICA for artifact removal as the
> components are removed form the data leaving you with all (or as many as
> possible) epochs to analyse.
> 3. What if after the second ICA I am still left with eye movements i.e.
> can't clearly identify which component is responsible for the eye
> Have I misunderstood something here? Please help,
> Dr Mary-Jane Budd
> Senior Research Officer
> Department of Psychology
> University of Essex
> Wivenhoe Park
> Colchester CO4 3SQ
> Room 4.726
> 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...
More information about the eeglablist