[Eeglablist] Why most of good 'brain' ICs are 'dipolar' with show 'red'-centerd scalp topos, although 2/3 of the cortex is in sulci?

Tue Dec 12 10:50:45 PST 2023

Hi Arno,

Thank you for your comment.

> Do you get the same results (majority of positively skewed, radial
equivalent source ICA components) if you initialize ICA with only
negatively skewed topography. In other words, can ICA override its
initialization?

If you start ICA with x-1 to all scalp topos, in theory the polarity will
be all flipped. But I haven't confirmed it yet in simulation.

> ICA is based on the probability density distribution of the activity. I
am unclear how the skewness of the distribution of the inverse weights
(scalp topography) of ICA is relevant to the algorithm. The skewness of the
ICA components activity might be though.

No, the skewness of the scalp topos. It was Scott's idea as mentioned in
the acknowledgements.

> I feel that your interpretation of ICA component not being present in
gyri might be premature.

You mean sulci.

> It would require that you assess the percentage of radially oriented
brain regions in a realistic cortical sheet and then compare it with the
distributions of radially oriented dipoles.

Figure 5 shows a summary that answers to your questions pretty well.

> I also think that an alternative hypothesis would be that the cortex is
uniformly active. However, ICA components might represent relatively large
patches of cortex that encompass both gyri and radial-oriented cortex --
making the dipoles mostly radially oriented.

Right. If you accept the large-patch hypothesis, that possibility is also
available.

> With a careful model based on this assumption, you might be able to
estimate the cortex patch size that ICA components represent.

First of all, we should learn how to make an effective use of the dipole
moments to ask those questions. I performed the patch-size estimation for a
SfN 2019 poster. But it was unsatisfactory because I realized ICA is a
statistical model after all.
The absolute amplitudes with unit is the only connection to the physics of
EEG, but our ICA analysis seems long abandoned it. We can easily give it an
upper and lower bounds from theory, and refine it based on empirical
observations.

Makoto

On Tue, Dec 12, 2023 at 1:20 PM Arnaud Delorme via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> These are interesting results Makoto.
>
> I would be interested:
> - Do you get the same results (majority of positively skewed, radial
> equivalent source ICA components) if you initialize ICA with only
> negatively skewed topography. In other words, can ICA override its
> initialization?
>
> - ICA is based on the probability density distribution of the activity. I
> am unclear how the skewness of the distribution of the inverse weights
> (scalp topography) of ICA is relevant to the algorithm. The skewness of the
> ICA components activity might be though.
>
> - I feel that your interpretation of ICA component not being present in
> gyri might be premature. It would require that you assess the percentage of
> radially oriented brain regions in a realistic cortical sheet and then
> compare it with the distributions of radially oriented dipoles. I also
> think that an alternative hypothesis would be that the cortex is uniformly
> active. However, ICA components might represent relatively large patches of
> cortex that encompass both gyri and radial-oriented cortex -- making the
> dipoles mostly radially oriented. With a careful model based on this
> assumption, you might be able to estimate the cortex patch size that ICA
> components represent.
>
> Arno
>
> > On Dec 11, 2023, at 11:53 PM, Pål Gunnar Larsson via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
> >
> > Just want to add. In rats about 50% of all fiber going in and out of the
> cortex are connected to the thalamus. In humans connections are about 2%,
> according to Nunez. Hence, we should be very careful when you try to
> extrapolate from animal research to humans.
> >
> > Pål G. Larsson
> >
> > Ikke sensitiv
> >
> >
> > -----Opprinnelig melding-----
> > Fra: eeglablist <eeglablist-bounces at sccn.ucsd.edu> På vegne av Makoto
> Miyakoshi via eeglablist
> > Sendt: 11. desember 2023 19:06
> > Til: EEGLAB List <eeglablist at sccn.ucsd.edu>
> > Emne: Re: [Eeglablist] Why most of good 'brain' ICs are 'dipolar' with
> show 'red'-centerd scalp topos, although 2/3 of the cortex is in sulci?
> >
> > Hi Scott,
> >
> >> "How are LFP signals across each of these gyrii synchronized across
> >> the
> > dataset?"
> >
> > The answer is not so special. The synchrony is achieved via
> thalamo-cortical loops.
> > In the following Wiki article, I linked to my presentation at an NIH
> summer seminar in which I showed multiple evidence that cortical synchrony
> and coupling is controlled by thalamus.
> >
> https://sccn.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline#Two_presentations_at_a_seminar:_EEG_preprocessing_and_generative_mechanism_.28For_240.2C000_page_views.2C_09.2F21.2F2023_added.2C_10.2F18.2F2023_updated.29
> >
> > So, when thalamus makes different cortical regions to fire together,
> then you see the synchronous activity. That's it. The distant cortical
> regions do not have to be directly connected via each neuron's lateral
> branches (which does exist, but the conduction speed is very slow compared
> with that of white matter) Note this is not a one-way 'imposing' the rhythm
> from the thalamus to the cortex, like the historical 'pace-maker'
> hypothesis by Andersen and Andersen (1967) but it is a bi-directional
> interaction.
> >
> > One fact that might help you see the situation is that only 1% of
> neurons need to be synchronized to form 95% of the amplitude of the
> observed signal according to Hari (1997).
> >
> > Also, it might also help to remember that there is no general guarantee
> that an EEG source is stationary and localizable. See Izhikevich's
> classical simulation.
> >
> https://urldefense.com/v3/__https://www.izhikevich.org/publications/large-scale_model_of_human_brain.htm__;!!Mih3wA!FLAtrGcNSxCRup1LcnfgWIJkdfC5HMOr_rPujEIjdmGug69GeOA7PxjXg5NqsRrbtx3VtKZRJhKrQ9HIAFxdjg0n5Fg$
> > There is an established principle of functional brain mapping but it is
> a product of statistical processing such as (heavy) averaging. ICA model is
> the same, hence it is stationary across time. The small and localizable EEG
> source is heavily a statistical concept. The actual ongoing EEG is
> stochastic, dynamic, and diffuse. When we see ICA results, therefore, we
> should distinguish properties of the filter from properties of data.
> >
> > Makoto
> >
> >
> > On Mon, Dec 11, 2023 at 11:42 AM Scott Makeig <smakeig at gmail.com> wrote:
> >
> >> Makoto -
> >>
> >> When you repeat the claim that EEG sources 'found' by ICA
> >> decomposition must be at least several adjacent gyrii in size, you
> >> fail to ask, "How are LFP signals across each of these gyrii
> synchronized across the dataset?"
> >> Doesn't this require some physiological basis, and if so, what is it??
> >>
> >> Scott Makeig
> >>
> >> On Mon, Dec 11, 2023 at 11:17 AM Makoto Miyakoshi via eeglablist <
> >> eeglablist at sccn.ucsd.edu> wrote:
> >>
> >>> Hello EEGLAB list,
> >>>
> >>> For those who have wondered so, here are my answers.
> >>> I asked two questions:
> >>>
> >>> (1) Why do good 'brain' ICs show dipolar scalp topos although 2/3 of
> >>> the cortex is in sulci?
> >>> (2) Why do these dipolar IC scalp topos show red (positive) centers?
> >>>
> >>> The answer was published a few days ago.
> >>>
> >>>
> >>> https://urldefense.com/v3/__https://onlinelibrary.wiley.com/doi/10.10
> >>> 02/hbm.26540__;!!Mih3wA!FPOThEiX2hsD7TJBq7WyhlV8v6HSkTe_swsBEoB2RM-Bh
> >>> -BGerduzZBnmEtDBamyosThbqv9Xrc1gGPSmdm52LpO7jM$
> >>>
> >>> The answer to (1): It is because scalp-recorded EEG is insensitive to
> >>> sulcal sources compared with gyral sources. This finding justifies
> >>> the use of lissencephalic (i.e. no sulci) brain model proposed in
> >>> Electric Fields of the Brain (Nunez and Srinivasan, 2006) together
> with Spline Laplacian.
> >>> This also supports the view that the major source of scalp-recordable
> >>> EEG is pretty broad (minimum 6.45 cm^2) which requires a continuum of
> >>> multiple gyral crowns.
> >>>
> >>> I did not write it in the paper, but the result basically refutes the
> >>> claim that ICA is a high-resolution EEG spatial filter because the
> >>> result confirms that ICA is mostly blind to 2/3 of the cortex. In
> >>> fact, it seems ICA results are always dominated by high-power,
> >>> low-frequency, and very broad sources. I will publish this view in
> >>> the near future.
> >>>
> >>> The answer to (2): It is because EEGLAB's ICA sets the initial topos
> >>> of all ICs red centered (i.e. positive dominant). Thus, unless
> >>> necessary, the algorithm does not flip the polarities.
> >>>
> >>> Now you wonder--when does the ICA algorithm flip the polarity to
> >>> produce 'blue' centered (i.e. negative dominant) ICs? I found that
> >>> those blue-centered ICs tend to show poor physiological validity with
> >>> large index numbers. A known clear exception for this rule is ICs
> >>> localized for the motor cortex.
> >>>
> >>> People use ICA to clean EEG. I use EEG to glean ICA, which is more fun.
> >>>
> >>> Makoto
> >>> _______________________________________________
> >>> 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
> > _______________________________________________
> > 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