[Eeglablist] ICA preprocessing and bipolar montage configuration
Makoto Miyakoshi
mmiyakoshi at ucsd.edu
Wed Jun 1 20:12:35 PDT 2022
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
>
>
More information about the eeglablist
mailing list