[Eeglablist] Rereference

Mon Oct 6 13:09:23 PDT 2014

Thank you Iman. Did I tell you this page too?
http://sccn.ucsd.edu/wiki/Linear_Representations_and_Basis_Vectors#Singular_Value_Decomposition_.28SVD.29

However, the wikipage linked above does not explain what's good about doing
the re-referencing as a pre-ICA preprocess. Hence I was happy to hear the
clear explanation for the reason.

Simon, I appreciate your honest opinion. I agree that the conclusion may
look damn, but I am happy to find it rather simple. After all, that's the
only thing ICA users want to care.

Jason, I agree to this:

> I would still of course suggest removing bad channels first for practical
reasons.

and I would say this is an optimism;

> However the change is just a linear addition of one noise direction along
the channel, which can be easily undone by ICA, at the cost of using one of
the ICA sources to contain this noise channel.

It would only hold for the cases of eye-blink, EOG, and EMG.
Artifacts with very high level amplitude (>10^4 uV), channels with
continuous zeros, or saturated channels can destroy ICA results or
terminates the converging process. We should protect ICA from these
artifacts rather than letting it handle them to fail (maybe this is a
different topic from the average referencing.)

Jason, I suggest we update the wiki page about the average referencing,
since this is a recurring topic on the list... if you don't mind I'll edit
your replies and prepare the draft for your check.

Makoto

On Sun, Oct 5, 2014 at 9:36 PM, Iman M.Rezazadeh <irezazadeh at ucdavis.edu>
wrote:

> I would like to suggest you read also the following archive :
>
> http://sccn.ucsd.edu/pipermail/eeglablist/2014/007810.html
>
>
>
>
>
> *From:* eeglablist-bounces at sccn.ucsd.edu [mailto:
> eeglablist-bounces at sccn.ucsd.edu] *On Behalf Of *Jason Palmer
> *Sent:* Sunday, October 5, 2014 5:33 PM
> *To:* 'Simon Finnigan'; mmiyakoshi at ucsd.edu
>
> *Cc:* 'eeglablist'; 'Eric HG'
> *Subject:* Re: [Eeglablist] Rereference
>
>
>
> Dear Eric, Simon, et al.,
>
>
>
> This issue has been discussed several times before and you can check the
> archives for more discussion. I have argued before, mainly from a linear
> algebra point of view, that any referencing using a linear combination some
> of the other channels is essentially projecting the data onto an space of
> dimension (nchan-1), where sources along a particular direction become
> invisible. In the case of average reference, this direction is the source
> that equally increases or decreases all the channel potentials by the same
> amount, i.e. [1 1 1 1 …. 1], and in consequence, all the learned ICA maps
> are orthogonal to this direction, and any sources along that direction are
> lost. But if there is any part of the source map that is not constant, this
> part will be retained and can be extracted by ICA.
>
>
>
> Using average reference makes all ICA sources have zero total potential. I
> have argued before that zero total potential makes sense for brain sources
> assuming approximately complete surrounding of head by electrode array, by
> a charge conservation argument. This will of course not hold exactly in
> practice, and there will be some shunting, so the learned ICA maps must be
> assumed to have an unknown constant potential added to each electrode for
> localization purposes.
>
>
>
> What I said to Makoto was that the reference didn’t really matter
> ultimately as long as you choose a combination of channels that cannot
> likely represent a real source. I suggest average reference because this
> makes all the maps sum to zero, clearly indicating
> dipolarity/directionality by having negative (blue) and positive (red)
> values in each map, rather than having maps that have dark blue background
> and light blue source “blob” (or dark red and light red). As I said, there
> must be assumed to be an unknown constant added to each map. My natural
> default for display purposes would be the total potential, and using
> average reference does this for you “automatically” with little risk of
> losing any sources of interest.
>
>
>
> People often worry that including a bad channel in the average reference
> will contaminate everything by adding noise to every channel. However the
> change is just a linear addition of one noise direction along the channel,
> which can be easily undone by ICA, at the cost of using one of the ICA
> sources to contain this noise channel. I would still of course suggest
> removing bad channels first for practical reasons.
>
>
>
> Best,
>
> Jason
>
>
>
>
>
> *From:* eeglablist-bounces at sccn.ucsd.edu [
> mailto:eeglablist-bounces at sccn.ucsd.edu <eeglablist-bounces at sccn.ucsd.edu>]
> *On Behalf Of *Simon Finnigan
> *Sent:* Saturday, October 04, 2014 2:51 AM
> *To:* mmiyakoshi at ucsd.edu
> *Cc:* eeglablist; Eric HG
> *Subject:* Re: [Eeglablist] Rereference
>
>
>
> Hi All
>
>
>
> Firstly I must say that as a scientist, I find the rationale from Jason
> via Makoto to be un-scientific and unsatisfactory.
>
>
>
> Second - This will only partly answer your question I think, but here
> goes: it all depends on how you're analysing your data [what you're looking
>  at/for] and particularly, whether or not scalp topographies are of
> interest to you - but in general and in brief, the consensus seems to be
> that common-average referencing is preferable over mastoid referencing, at
> least - and probably other references sites too [e.g. vertex; nasion].
> [Also are your data referenced to one or both mastoids? and if both - were
> they linked, or averaged together after recording?]
>
>
>
> My 2002 paper in 'Neuropsychologia' addressed this issue scientifically,
> at least in part - e.g. see final figure [albeit we weren't doing ICA back
> then either!].
>
>
>
> Regards, Simon
>
>
>
>
> On 04/10/2014, at 8:21, Makoto Miyakoshi <mmiyakoshi at ucsd.edu> wrote:
>
> Here is update.
>
> I asked the question to Jason. He told me that the main reason avg
> reference is useful is that it returns brain maps that look like nice
> localized sources inside the head with red blob and slight negative
> (aqua/green) background. This makes sense.
>
>
>
> Makoto
>
>
>
> On Thu, Oct 2, 2014 at 10:00 AM, Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
> wrote:
>
> Dear Eric,
>
>
>
> I have the same question for long time. In my case, because I run ICA and
> remain in IC analysis and almost never come back to channels, I really
> wonder if I need re-referencing at all.
>
>
>
> I've heard Jason Palmer, who wrote AMICA for EEGLAB, say average reference
> before ICA is necessary, but he did not give me a clear explanation for
> that. Scott wanted to know it too, but we were interrupted at that time.
> Next time I'll ask it to him (when I ever see him in the lab...)
>
>
>
> Makoto
>
>
>
> On Tue, Sep 30, 2014 at 2:44 PM, Eric HG <erichg2013 at gmail.com> wrote:
>
> Dear list,
>
>
>
> I have read the tutorial on the site and it says: "Converting data,
> before analysis, from fixed or common reference (for example, from a common
> earlobe or other channel reference) to 'average reference' is advocated by
> some researchers, particularly when the electrode montage covers nearly the
> whole head (as for some high-density recording systems)."
>
>
>
> I have data collected with a mastoid reference and I was wondering in
> which cases it would be considered best to rereference to the average? Are
> there any specific papers about rereferencing that anyone would recommend?
>
>
>
> Regards,
>
>
>
> Eric
>
>
>
> _______________________________________________
> 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
>
>
>
>
>
> --
>
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>
> _______________________________________________
> 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/20141006/dcbee883/attachment.html>