[Eeglablist] Re-reference after ICA cleaning for BioSemi data

[Makoto Miyakoshi]

Dear Ben,

> 1) Given that BioSemi data is collected essentially "reference-free", is
it OK to perform ICA before re-referencing?

That's my understanding, since common-mode noise you reject by
re-referencing is basically a linear operation which ICA should not be
affected. HOWEVER, if your common-mode noise is huge, I recommend you
average-reference before ICA--it may help (or not, I don't know but should
not hurt).

> 2) After using ICA to remove eye blink components and project back to
sensor space, can I re-reference the cleaned data directly?  Or do I need
to first re-reference the original data AND the unmixing matrix from the
ICA before removing the bad components?

I discussed it with my super smart colleague Jason Palmer. He strongly
recommended you average-reference BEFORE ICA. The reason for it is that it
realizes zero-sum assumption that dipole model has... which sounded quite
trivial to me but again it does not hurt! If you re-reference before ICA
you want to explicitly adjust the rank by pca option or you want to reject
one channel (I always recommend this solution since simple).

Makoto

On Thu, Jan 7, 2016 at 1:04 PM, Benjamin Skerritt-Davis <bsd at jhu.edu> wrote:

> Hi everyone,
>
> I need some advice on cleaning BioSemi EEG data using ICA. I have data
> with 38 channels, and I would like to be able to analyze the data using
> different re-reference channels after ICA cleaning (e.g., linked mastoids,
> nose). My intuition is to run ICA on all 38 channels (before
> re-referencing) so the data is full rank, project out any "bad" components,
> then re-reference the cleaned data.  A couple of questions:
>
> 1) Given that BioSemi data is collected essentially "reference-free", is
> it OK to perform ICA before re-referencing?
>
> 2) After using ICA to remove eye blink components and project back to
> sensor space, can I re-reference the cleaned data directly?  Or do I need
> to first re-reference the original data AND the unmixing matrix from the
> ICA before removing the bad components?
>
>
> I've found conflicting answers to similar questions online, so I'm hoping
> for consensus this time around. =)
>
> Thanks for the help,
> -Ben
>
> _______________________________________________
> 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
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20160111/665eaf27/attachment.html>