[Eeglablist] microstate analysis
Scott Makeig
smakeig at gmail.com
Wed May 8 14:48:16 PDT 2024
All -
I haven't dived deeply into 'microstate analysis', but find the title
distasteful - When a certain portion of the scalp is net positive or
negative, is one to believe that represents (in general) a certain 'state'
of the brain or mind? As Kevin nicely expressed, any one channel sums
potentials emanating from a particular, broad (though not contiguous) swath
of cortex. Thus, a certain scalp pattern may appear during a (technically,
near infinite) collection of (very likely unrelated) cortical potential
maps.
However as Fiorenzo Artoni pointed out to me, 'microstate' decomposition is
in essence a method (in this way not unlike PCA) for reducing the
dimensionality of multi-channel data while retaining a relatively large
percentage of variance. Like PCA and (any) other linear decompositions,
data in a 'microstate' data subspace can be used for any kind of
statistical analysis, BCI, etc. - it should retain at least some of the
information about brain dynamics available in the raw data.
To try to use it for purposes of brain imaging seems at best unpromising,
when other methods based on more physiologically plausible assumptions are
available.
My thoughts ...
Scott Makeig
p.s. I would hope that users would first apply ICA decomposition and then
at least separate out eye blinks, line noise, muscle activities so as to
identify a 'more possibly brain' data subspace on which to apply their
'microstate decomposition.'
p.p.s. Of course, in a particular task paradigm, for example the period of
the Late Positive Complex ('p300') in the 'target stimulus
presentation-locked' ERP, more positivity in a particular paradigm, subject
cohort, etc. may indicate something about e.g. sensory expectation
(depending on paradigm, subject cohort, etc.). But here, too, the lack of
anatomic specificity limits both the information content and the
physiological interpretation. Our ICA decompositions showed early on that
'the P300' (as reported in 1000s of ERP papers) can (given enough channels
and other favorable data characteristics) be separated using ICA
decomposition into at least 9 clusters of strongly localizable cortical
sources, whose relative contributions to 'the' scalp positivity varied from
trial to trial, subject to subject, and across subject groups.
Interpretation of these differences in general correlated with fMRI and
other neuroscientific evidence.
On Wed, May 8, 2024 at 4:56 PM Ramesh Srinivasan via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:
> Kevin
>
> Your description here nicely expressed my view of this for a number of
> years. However my views have changed, and I believe this is a useful way
> to think about the data and the brain.
>
> In recent years a number of labs (including mine) have been interested in
> deciding EEG at the.trial level to investigate perceptual and cognitive
> states by making predictions. This approach is a bit inspired by BCI folks
> who face the challenge of operating devices from single trials of EEG.
>
> I found intellectual appeal in this. If you think about it neurons decide
> the output of other neurons and EEG reflects this. So we should learn to do
> this and maybe we get some insight on functional signals.
>
> Recently I have come to appreciate that all of the people looking at trial
> level EEG predictors are actually interested in microstates that encode
> perceptual and cognitive states. We just haven't been calling it that.
> But what we are doing overlaps with that approach.
>
>
> Ramesh Srinivasan
> Professor
> Cognitive Sciences
> Biomedical Engineering
>
> On Mon, May 6, 2024, 10:04 AM Kevin Spencer via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
>
> > Hello EEGers,
> >
> > Another couple of posts brought up the issue of microstate analysis,
> which
> > has been bothering me recently. I'm not sure if I'm missing something,
> but
> > the basic approach with ERPs, clustering topographies across the epoch
> into
> > discrete "states", seems to ignore the well-established fact that the
> > topography of the EEG across the scalp at a given time point represents
> the
> > sum of numerous spatially and temporally overlapping activity patterns.
> > That's why people originally applied PCA to ERP data, to disentangle
> these
> > overlapping patterns. A classic example is that the "late positive
> complex"
> > consists of several ERP components that overlap each other in time and
> > space (e.g., P3a, P3b, classic Slow Wave, etc.).
> >
> > If you apply some type of cluster analysis to the topographies of epochs
> > of EEG single trials or ERPs, you are going to get clusters that
> represent
> > maxima in global field power, but which say nothing about the
> multiplicity
> > of activity patterns that contribute to these clusters. In contrast, if
> you
> > apply say ICA to the data, you decompose the data into distinct activity
> > patterns that do overlap in various dimensions. So I don't see what is so
> > useful about microstate analysis. But perhaps I'm missing something. I'd
> be
> > interested in what people who know this method better than I do have to
> say.
> >
> > There is also the whole issue of what a brain "state" is. I'm not aware
> of
> > any reason to think that a scalp topography that lasts for X ms
> constitutes
> > a brain state. But again, maybe I'm not aware of research that supports
> > this idea.
> >
> > Kevin
> >
> >
> >
> ---------------------------------------------------------------------------------
> > Kevin M. Spencer, Ph.D.
> > Research Health Scientist, VA Boston Healthcare System
> > Associate Professor of Psychiatry, Harvard Medical School
> >
> >
> ---------------------------------------------------------------------------------
> > _______________________________________________
> > Eeglablist page:
> >
> https://urldefense.com/v3/__http://sccn.ucsd.edu/eeglab/eeglabmail.html__;!!CzAuKJ42GuquVTTmVmPViYEvSg!PAPQu90khb16mTEftF0J01W4PZwcI63VHdDKg_r0aqVWTtAvn-bBrAMaLhk2ApbZPUvcXJV_rPPsKgKRUfBnMFAq$
> > 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
--
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
More information about the eeglablist
mailing list