[Eeglablist] ICA preprocessing and bipolar montage configuration
Velu Prabhakar Kumaravel
velu.kumaravel at unitn.it
Thu Jun 2 03:12:02 PDT 2022
Thank you very much for a valuable discussion.
I confirm Makoto's last message about realistic bipolar reference - When I
tried converting data back to unipolar reference, I realized I do not have
enough information as explained by Makoto.
If it is useful to add, I am analyzing the CHB-MIT database
<https://urldefense.proofpoint.com/v2/url?u=https-3A__physionet.org_content_chbmit_1.0.0_&d=DwIFaQ&c=-35OiAkTchMrZOngvJPOeA&r=kB5f6DjXkuOQpM1bq5OFA9kKiQyNm1p6x6e36h3EglE&m=rITCkKk2XaPrAvKumpGt1_L1Kzs0LMGlIexTW6slq-qap81a8u2zwqbK_9iIE5Fr&s=3_pfhPQHzpMaG_m95NJccLSFVDQoY9qLgrwLCStUSDc&e= >. Thank you once again.
Best,
Velu
On Thu, 2 Jun 2022 at 05:15, Makoto Miyakoshi via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:
> Jason, your example uses redundant bipolar referencing, which is A-B, B-C,
> C-D as I explained above. This can be confirmed in your example below (I
> changed your B to M because I use A B C D for electrode indices)
>
> > A bipolar channel matrix looks like:
> >
> > M = [ 1 -1 0 0 0 0 0 …. ;
> >
> > 0 1 -1 0 0 0 0 …. ;
> >
> > 0 0 1 -1 0 0 0 … ;
>
> Note the rank of this data is nbchan-1, which is the same as the unipolar
> recorded EEG data including the initial reference. In other words, your
> example does not differ from unipolarly recorded data!
>
> But I'm talking about the case in which only A-B and C-D pairs are
> available without B-C. The reason why I assume this is because this is the
> reality of bipolar reference. In this case, M is
>
> M = [ 1 -1 0 0 0 0 0 …. ;
> 0 0 1 -1 0 0 0 … ;
> 0 0 0 0 1 -1 0 ...;
>
> M is not invertible.
>
> > It should be possible to define a square, invertible bipolar matrix
> (which I called B, but is called M here).
>
> I want to first confirm if your explanation is still valid for my
> 'realistic bipolar reference M' scenario.
>
> Makoto
>
> On Wed, Jun 1, 2022 at 7:13 PM Jason Palmer <japalmer29 at gmail.com> wrote:
>
> > Hi Makoto,
> >
> >
> >
> > It should be possible to define a square, invertible bipolar matrix
> (which
> > I called B, but is called M here). Then I don’t think there needs to be a
> > caveat about the interpretation.
> >
> >
> >
> > Jason
> >
> >
> >
> > *From:* Makoto Miyakoshi [mailto:mmiyakoshi at ucsd.edu]
> > *Sent:* Wednesday, June 1, 2022 11:12 AM
> > *To:* Scott Makeig <smakeig at gmail.com>; Jason Palmer <
> japalmer29 at gmail.com
> > >
> > *Cc:* EEGLAB List <eeglablist at sccn.ucsd.edu>
> > *Subject:* Re: [Eeglablist] ICA preprocessing and bipolar montage
> > configuration
> >
> >
> >
> > Dear Scott and Jason,
> >
> >
> >
> > See this comment from one of the discussions on the list from 2017,
> > particularly the last highlighted part.
> >
> > Scott, we discussed this issue during tea time. Did you reach a
> > different conclusion? Do we want to correct this conclusion?
> >
> >
> >
> > Makoto
> >
> >
> >
> > %%%%%%%%%%%%%%%%%
> >
> > https://sccn.ucsd.edu/pipermail/eeglablist/2017/012597.html
> >
> >
> >
> > [Eeglablist] convert EEG montage from bipolar to unipolar
> > Makoto Miyakoshi mmiyakoshi at ucsd.eduThu May 25 12:35:12 PDT 2017
> >
> >
> >
> > Dear colleagues,
> >
> > Update--
> >
> > I discussed this method with the colleague who taught me about this trick
> >
> > because I got an inquiry about it off the list.
> >
> > He said that he would use it only to draw a scalp topography, and
> >
> > performing signal processing using the 'recovered' full-channel signal is
> >
> > not recommended.
> >
> >
> >
> > Again, let X be the (single-channel referenced) original EEG data and Xb
> >
> > the bipolar-montage version of it.
> >
> > Using a bipolar-referencing transform matrix M, the relation of X and Xb
> >
> > can be written as
> >
> >
> >
> > Xb = M * X
> >
> >
> >
> > Suppose X has 6 channels F1, F2, C1, C2, P1, P2, and bipolar-referencing
> >
> > was done with F1-F2, C1-C2, P1-P2.
> >
> > Then, Xb is 3*t, M is 3*6, and X is 6*t (t is time).
> >
> > The matrix M is
> >
> >
> >
> > 1 -1 0 0 0 0
> >
> > 0 0 1 -1 0 0
> >
> > 0 0 0 0 1 -1
> >
> >
> >
> > (sorry but I don't have code)
> >
> > To compute M^-1, one should use pinv() (i.e. pseudo-inverse) and the
> >
> > recovered full-channel data are NOT full-ranked (and this is NOT the only
> >
> > solution)
> >
> > To recover full channel data, you also need to have something like F1-C1,
> >
> > C2-P1, P2-F2 so that the matrix is full-ranked and square, but such data
> >
> > are not usually available (as far as I know, Paul Sajda presented such a
> >
> > complicated reference system to address high-amplitude artifact in
> >
> > simultaneous fMRI-EEG recording).
> >
> >
> >
> > So the recommended use of this solution is just to draw scalp topography
> >
> > for convenience.
> >
> >
> >
> > Accordingly, let me correct my previous statement.
> >
> >
> >
> > >* Thus, by multiplying the inverse matrix of the known
> bipolar-referencing*
> >
> > transform matrix M, you obtain the original signal.
> >
> >
> >
> > This is true ONLY IF you have redundantly referenced channels (which is
> >
> > very rare). Otherwise, it does NOT properly convert standard bipolar
> >
> > montage to a single-referenced data (because M is not square), therefore
> >
> > analyzing the 'recovered' data should be limited to specific purposes
> only.
> >
> >
> >
> > Sorry if my previous writing was misleading.
> >
> >
> >
> > Makoto
> >
> >
> >
> > On Wed, Jun 1, 2022 at 10:29 AM Scott Makeig <smakeig at gmail.com> wrote:
> >
> > Jason -
> >
> > Thanks - exact, as usual for you ... I am looking to try it - then put it
> > into EEGLAB.
> >
> > Scott
> >
> > On Wed, Jun 1, 2022 at 1:13 PM Jason Palmer <japalmer29 at gmail.com>
> wrote:
> >
> > > ICA will try to produce independent sources (the s below) regardless of
> > > how they are linearly mixed.
> > >
> > >
> > >
> > > A bipolar channel matrix looks like:
> > >
> > >
> > >
> > > B = [ 1 -1 0 0 0 0 0 …. ;
> > >
> > > 0 1 -1 0 0 0 0 …. ;
> > >
> > > 0 0 1 -1 0 0 0 … ;
> > >
> > > ...
> > >
> > > ];
> > >
> > >
> > >
> > > Or similar, depending on how you define the bipolar channels.
> > >
> > >
> > >
> > > And the raw data x, after bipolar channel extraction is B*x = B*A*s,
> > where
> > > A is mixing matrix topo maps, and s is the source vector. If you run
> ICA
> > on
> > > B*x, then you get a decomposition with a mixing matrix M. This is not
> > > necessarily interpretable with topoplot. To get the actual topo maps,
> you
> > > need to invert the bipolar matrix:
> > >
> > >
> > >
> > > B*A = M => A = pinv(B)*M
> > >
> > >
> > >
> > > where again, B is the custom bipolar channel difference matrix, and M
> is
> > > the EEG.icawinv after running ICA on B*x (where x is EEG.data). The
> > columns
> > > of this A matrix should be as usual. Using the bipolar differences
> might
> > > remove more far field noise and components vs usual raw channels.
> > >
> > >
> > >
> > > Jason
> > >
> > >
> > >
> > > *From:* Scott Makeig [mailto:smakeig at gmail.com]
> > > *Sent:* Wednesday, June 1, 2022 10:01 AM
> > > *To:* Jason Palmer <japalmer29 at gmail.com>
> > > *Cc:* Velu Prabhakar Kumaravel <velu.kumaravel at unitn.it>; EEGLAB List
> <
> > > eeglablist at sccn.ucsd.edu>
> > > *Subject:* Re: [Eeglablist] ICA preprocessing and bipolar montage
> > > configuration
> > >
> > >
> > >
> > > Jason - But the mapping problem remains, no? Will ICA make the IC maps
> > > (including somehow the bipolar channel) smooth even when the data
> > contains
> > > a bipolar channel? What about this argument: Making such a dataset
> > > zero-mean does not necessarily mean that the difference between the
> > common
> > > reference and each electrode in the bipolar channel is 0. If the
> sources
> > > were [equally everywhere], then this might be the case, but in actual
> > fact,
> > > is it?
> > >
> > >
> > >
> > > Scott
> > >
> > >
> > >
> > > On Wed, Jun 1, 2022 at 12:07 PM Jason Palmer <japalmer29 at gmail.com>
> > wrote:
> > >
> > > A bipolar montage is basically just a linear transformation, and
> assuming
> > > you use at most nbchan number of bipolar channels, it is invertible.
> > >
> > > x = A*s
> > > B*x = B*A*s = M*s
> > > A = inv(B) * M
> > >
> > > Where B is the bipolar montage with 1 -1 in the columns corresponding
> to
> > > the
> > > bipolar difference for each row, and M is the EEG.icawinv after running
> > ica
> > > on the bipolar transformed data.
> > >
> > > Jason
> > >
> > > -----Original Message-----
> > > From: eeglablist [mailto:eeglablist-bounces at sccn.ucsd.edu] On Behalf
> Of
> > > Scott Makeig
> > > Sent: Wednesday, June 1, 2022 8:52 AM
> > > To: Velu Prabhakar Kumaravel <velu.kumaravel at unitn.it>
> > > Cc: EEGLAB List <eeglablist at sccn.ucsd.edu>
> > > Subject: Re: [Eeglablist] ICA preprocessing and bipolar montage
> > > configuration
> > >
> > > Velu -
> > >
> > > ICA is a linear decomposition, so should be able to decompose bipolar
> and
> > > common-reference channels in the same session together. However, the
> > > bipolar
> > > channels will be 'floating' with respect to the common-reference
> > channels.
> > > This might interfere with the decomposition (I have no practical
> > experience
> > > here) - but even if not it will mean that the ICA scalp maps should not
> > be
> > > plotted to include the bipolar channels. Here I imagine you might try
> to
> > > find fixed offsets representing a 'standing difference'
> > > between each bipolar channel and the common reference channel that
> would
> > > produce max smooth IC maps -- but are these differences reliably
> > > stationary? I'd be interested to see a result of attempting this ...
> > >
> > > Scott Makeig
> > >
> > > On Wed, Jun 1, 2022 at 11:42 AM Velu Prabhakar Kumaravel <
> > > velu.kumaravel at unitn.it> wrote:
> > >
> > > > Dear EEGLABers,
> > > >
> > > > Does anyone know the effects of ICA preprocessing on EEG acquired
> > > > using bipolar configuration?
> > > > I tried on a few datasets and it looks like the decomposition is not
> > > > effective. Classifying using ICLabel results in more number of
> "Other"
> > > > category.
> > > >
> > > > Could someone provide insights on this?
> > > >
> > > > Best regards,
> > > >
> > > > Velu Prabhakar Kumaravel, Ph.D. Student Center for Mind/Brain
> > > > Sciences, University of Trento, Italy
> > > > _______________________________________________
> > > > 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
> > > _______________________________________________
> > > 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
> > >
> >
> >
> > --
> > 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
> > _______________________________________________
> > 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
> >
> >
> _______________________________________________
> 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
More information about the eeglablist
mailing list