[Eeglablist] coherence between subsets of electrodes

[Scott Makeig]

Magda -

The abstract you cite appears to me to have the right basic idea - but
applied in the wrong order.  Since coherence analysis is
nonlinear (involving signal cross-products) - and each scalp channel is a
(poor) spatial filter summing activity in distributed but similarly
oriented cortical territories - cross-products of such sums of source-area
signals (i.e, of scalp channels), involved in the coherence computation,
cannot be interpreted spatially.

The much better procedure is to perform spatial filtering on the channel
data, e.g. using ICA or other beamforming, to identify what I call
'effective sources' of the EEG signals, then localize the identified (or
therein approximated) spatial sources and compute coherence between them.
The coherence results can then be interpreted spatially.

The key insight is to realize that:
  (1) all EEG signals (or signal combinations) are spatial filters linearly
combining (summing) voltage fluctuations from many brain (plus non-brain)
sources,
  (2) single (or locally averaged) scalp channels are quite inefficient as
spatial filters themselves,
  (3) there are data-driven methods available for constructing precise
linear spatial filters from the raw channel signals whose source can be
brain-localized, and
  (4) that these spatially-filtered are just as 'real' EEG signals as scalp
channels.

Actually, I would say that the derived 'effective source' are more 'real'
than scalp channels, when they each are strongly dominated by activity from
just one brain (cortical) area.

Scott Makeig


On Mon, May 24, 2021 at 12:24 PM Magda Gawłowska via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> Dear All,
>
> I'm currently trying to figure out a way to compute NxN coherence in the
> electrode space. In short, I have 256 channel resting-state data (open and
> closed eyes) for two groups and I want to extract a coherence matrix, then
> run group x condition analysis.
>
> First of all, I wanted to figure out a way to compute the coherence not
> between the singular electrodes but between electrode groups. I thought
> about averaging the signal from the electrodes before computing coherence,
> but I think that in this case a simple solution is probably not the best
> one. I did find some data-driven approaches e.g. extracting functional
> units (based on this paper:
>
> https://urldefense.com/v3/__https://ieeexplore.ieee.org/abstract/document/4433991__;!!Mih3wA!TxzbfsYnXtDEVN3Rl16GNnMzoZCsQgqrmfz45ep1oqoO_QuQlfQAl-W3UE0QUlDxLroWRA$
> ). However, I get bug
> after bug and when I fix one problem, another appears. Up until now I've
> been able to figure out my scripts one way or another,  but this one drives
> me crazy.
>
> Does any of you can suggest a different approach to this kind of analysis?
> Or maybe some of you used the approach I mentioned and can guide me on how
> to make this method work?
>
> Second of all - assuming that the firsts step is done ;) and I'm am a happy
> owner of the coherence maps - how do you compare them? The only thing that
> comes to my mind is extracting some kind of graph index to do so... Do you
> have any ideas regarding this issue?
>
> Thanks in advance!
> Magda
> _______________________________________________
> 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