[Eeglablist] Open collaboration project: Multi-model Anatomical Grouping of Independent Components (MMAGIC)
Scott Makeig
smakeig at gmail.com
Fri Sep 18 09:19:57 PDT 2026
I didn't see mention in this thread of the primary problem facing nearly
all EEG source localization methods -- namely, determining the resistivity
of the skull to put into the electrical forward problem model used to
perform the inverse localization from the scalp map (e.g., from the IC
map). In 2013, Zeynep Akalin Acar's simulations
<https://urldefense.com/v3/__https://pubmed.ncbi.nlm.nih.gov/23355112/__;!!Mih3wA!EipENAca6HLV7GPzrkIRNLpmqblhZefs_Q65dTAUEKdGptI1A1_v0YeYAD2unu58hJMbRQAIZLeiQ1NzM6ky$ >confirmed that errors here can
move the estimated source depth inwards or outwards by up to 2 cm(!). This
matches what we see in dipole model source localization (e.g., using
dipfit() or NFT <https://urldefense.com/v3/__https://pubmed.ncbi.nlm.nih.gov/20457183/__;!!Mih3wA!EipENAca6HLV7GPzrkIRNLpmqblhZefs_Q65dTAUEKdGptI1A1_v0YeYAD2unu58hJMbRQAIZLeiQ7JBaKd-$ >), where the
estimated source location is inferior to cortical grey matter. While ground
truth values are difficult to obtain in living humans (and postmortem
measurements might also differ from in vivo), using our ICA
decomposition-based SCALE algorithm
<https://urldefense.com/v3/__https://pubmed.ncbi.nlm.nih.gov/26302675/__;!!Mih3wA!EipENAca6HLV7GPzrkIRNLpmqblhZefs_Q65dTAUEKdGptI1A1_v0YeYAD2unu58hJMbRQAIZLeiQ_fbITBO$ > consistently showed differences
in estimated skull conductance/resistivity of 3:1 in adult subjects. In our
first tests, SCALE estimates for the same subjects on different days (and
performing different tasks) varied much less than between-subject estimates
<https://urldefense.com/v3/__https://pubmed.ncbi.nlm.nih.gov/36086241/__;!!Mih3wA!EipENAca6HLV7GPzrkIRNLpmqblhZefs_Q65dTAUEKdGptI1A1_v0YeYAD2unu58hJMbRQAIZLeiQwMwTijr$ >, increasing our confidence in
SCALE accuracy. And -- using the SCALE-adjusted individual head models
(built using subject anatomical MR images) gave better-fitting distributed
cortical surface estimates than using the common template-value skull
conductance value.
If the electrical forward-problem head model is wrong in any way, any
source location estimates derived using it will be wrong to some extent (to
an extent depending on exactly what source is being estimated)...
EEG has long been embraced as a low-cost 'brain' dynamics measurement
system. This was okay when measurements were confined to how potentials
appear at the scalp. But actually addressing the real problem of better
understanding (cortical mesoscale) brain local field dynamics using
noninvasive scalp measurements poses the challenge of requiring an accurate
head model -- both its geometry (captured using MR) and its tissue
conductance values. Of the latter, skull conductance is known to vary most
widely between individuals.
Yahya Shirazi and student are now in touch with Zeynep to make the NFT
toolbox available and fully functional. I hope that this will include the
SCALE algorithm to give EEGLAB a high-resolution EEG source localization
procedure that can be employed routinely by users.
Scott Makeig
(note: Someone once pointed out to me that sweaty scalp tissue should
spread current more than drier scalp tissue - though I know of no actual
study of the effects of this on EEG source localization).
On Fri, Sep 18, 2026 at 12:06 AM Makoto Miyakoshi via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:
> Hi Cedric,
>
> I'd argue we should finish the 1/f project first :)
>
>
> Gin's paper serves as an interim report. She is one of our 1/f group.
>
> Cruz GE (2006) One slope, many causes: a multiscale framework for
> interpreting 1/f-like neural spectra
>
> https://urldefense.com/v3/__https://www.academia.edu/3071-4087/2/3/10.20935/AcadNeurosci8480__;!!Mih3wA!HVmlW9No06V_lAZ24_sp4qMfezgxfQOxhzlxG8ZcfDyo8i9LMG7ssI-REpHUhAYZF9dkpzrtkjp4R17-A_1Uvb-nmeM$
>
> She agreed to give us a lecture in Nov or Dec.
>
> I learned the cable theory and electrodiffusive
> neuron-extracelluler-glia (edNEG) model and pretty satisfied. This needs to
> be disseminated somehow. So I've been thinking about it.
>
> On the next steps, while we're at it, I'd love to add a comparison of ICA
> against GED as one of the decompositions feeding the ROI assignment. I've
> been using GED for frontal midline theta (FMT) following Zuure et al.
> (2020) and mike X Cohen's work on spatial filters. It is a guided
> filter: you hand it an S versus R covariance contrast and it returns the
> full set of components ranked by that contrast, of which you keep the top
> one or a few, so it needs a hypothesis in a way ICA does not. But where the
> contrast is well defined a priori, it's worth knowing whether the resulting
> component localizes better or worse than the ICs from the same data. It
> should also be cheap to test in this framework. If the core function takes
> a spatial pattern plus chanlocs, then a GED component's forward model,
> meaning the activation pattern rather than the filter weights, is the same
> kind of channel-space object as a column of EEG.icawinv, so it goes through
> the identical source pipeline with no extra machinery.
>
> We can imagine a situation in which ICA and GED can be reasonably compared.
> But a general comparison would be difficult.
> If you want to learn a smart way to use GED, check GEDAI and ZapLine. These
> are two excellent examples.
> Mike's NeuroImage paper on GED is also a nice text to learn it.
>
> Two small notes on your steps 1 and 2, since you asked for criticism. A
> center of mass over the whole eLORETA map can land in white matter or
> between two lobes when the solution is broad or bilateral
>
> That's right. We need something like spherical centroid or Frechet mean to
> stay on the surface of the cortex.
>
>
> so it may be worth reporting a spread measure alongside the centroid, or
> computing the centroid over a thresholded mass rather than the whole
> volume.
> And residual variance does not mean the same thing for a distributed model
> as for a single dipole, since the residual is set by the regularization and
> goes to zero as lambda does. It stays informative within one method with
> lambda fixed, but probably not across methods.
>
>
> We can't use 'residual variance' in the distributed source model, in the
> sense that it would not be meaningful since the degrees of freedom are
> too high. Am I correct?
>
> I initially thought of using some kind of spatial 'standard deviation' to
> express the blurriness or spatial spreading, which we can find for sure,
> but realized that anatomical ROI membership itself represents it. By the
> same token, we probably do not need to use the concept of centroid because
> a ROI which captures the distributed sources most is the center.
>
> Conceptually, by using both dipole fitting and a distributed source model,
> I want to give 'depth bounds' to an IC-decomponsed source; the lower
> (deeper) bound is dipole's depth due to depth bias, and the upper
> (shallower) bound is the depth of gray-matter constrained distributed
> source model. I predict that deeper-fit dipoles will show a
> broader-distributed source model for the case of qualified brain ICs.
>
> Makoto
>
>
> On Thu, Sep 17, 2026 at 4:48 PM Cedric Cannard via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
>
> > Hi Makoto and all,
> >
> > I'd argue we should finish the 1/f project first :)
> >
> > On the next steps, while we're at it, I'd love to add a comparison of ICA
> > against GED as one of the decompositions feeding the ROI assignment. I've
> > been using GED for frontal midline theta (FMT) following Zuure et al.
> > (2020) and mike X Cohen's work on spatial filters. It is a guided
> > filter: you hand it an S versus R covariance contrast and it returns the
> > full set of components ranked by that contrast, of which you keep the top
> > one or a few, so it needs a hypothesis in a way ICA does not. But where
> the
> > contrast is well defined a priori, it's worth knowing whether the
> resulting
> > component localizes better or worse than the ICs from the same data. It
> > should also be cheap to test in this framework. If the core function
> takes
> > a spatial pattern plus chanlocs, then a GED component's forward model,
> > meaning the activation pattern rather than the filter weights, is the
> same
> > kind of channel-space object as a column of EEG.icawinv, so it goes
> through
> > the identical source pipeline with no extra machinery.
> >
> > Two small notes on your steps 1 and 2, since you asked for criticism. A
> > center of mass over the whole eLORETA map can land in white matter or
> > between two lobes when the solution is broad or bilateral, so it may be
> > worth reporting a spread measure alongside the centroid, or computing the
> > centroid over a thresholded mass rather than the whole volume. And
> residual
> > variance does not mean the same thing for a distributed model as for a
> > single dipole, since the residual is set by the regularization and goes
> to
> > zero as lambda does. It stays informative within one method with lambda
> > fixed, but probably not across methods.
> >
> > Best,
> > Cedric
> >
> >
> > -------- Original Message --------
> > On Wednesday, 09/16/26 at 21:20 m za via eeglablist <
> > eeglablist at sccn.ucsd.edu> wrote:
> > Hi Sarvenaz,
> >
> > Thank you for your message! I’m very glad to connect with you and would
> be
> > happy to collaborate on the project.
> >
> > I think your experience with EEG/ERP data, validation, and statistical
> > evaluation could be very valuable for the project, and I’m looking
> forward
> > to working together.
> >
> > Best,
> > Marjan
> >
> > On Thu, 17 Sept 2026, 02:19 Sarvenaz Changizi via eeglablist, <
> > eeglablist at sccn.ucsd.edu> wrote:
> >
> > > Dear Makoto,
> > >
> > >
> > >
> > > Thank you very much for your detailed, clear, and encouraging response.
> > > The project sounds genuinely fascinating, and I would be very happy to
> > > contribute as an alpha user by applying the developed method to
> empirical
> > > EEG data, evaluating its outputs, and providing feedback throughout the
> > > development process. I would also be glad to contribute to the
> > validation,
> > > interpretation, documentation, and manuscript preparation wherever
> > useful.
> > >
> > >
> > >
> > > I have previous experience working with and contributing to the
> analysis
> > > of EEG/ERP data using EEGLAB. My research also involves designing
> > > experimental psychology tasks, and I am quite familiar with Python and
> > > programming-oriented research workflows. Therefore, I would be very
> > > interested in expanding my knowledge of the relevant dipole and
> > distributed
> > > source models and engaging with the technical aspects of the project.
> > >
> > >
> > >
> > > It was also a lovely surprise to hear that Marjan is based in Tehran—I
> am
> > > Iranian and originally from Tehran as well! So, although I would be
> > joining
> > > from Valencia, the team would actually be gaining another Tehran
> > > connection. :) I imagine that this shared background could make it
> > > especially easy and enjoyable for Marjan and me to communicate and
> > > collaborate.
> > >
> > >
> > >
> > > Could you please let me know what the best next step would be? Is there
> > > already a separate mailing list, group, shared repository, or
> > communication
> > > channel for the collaborators? I would be very happy if you could
> connect
> > > me with Marjan—perhaps by copying us both on an introductory email—and
> > let
> > > me know how the team plans to coordinate the work.
> > >
> > >
> > >
> > > Thank you again for taking the time to answer all my questions so
> > > thoroughly. I am genuinely excited about the possibility of joining the
> > > collaboration and contributing to the project.
> > >
> > >
> > >
> > > Best wishes,
> > > Sarvenaz
> > >
> > >
> > >
> > >
> > >
> > > >
> > > > On 16 Sep 2026, at 17:32, Makoto Miyakoshi via eeglablist <
> > > eeglablist at sccn.ucsd.edu> wrote:
> > > >
> > > > Hi Sarvenaz,
> > > >
> > > > Thank you for your questions. Here are my answers.
> Psychology/cognitive
> > > > science research skills would not be very helpful in the core
> > development
> > > > phase. It's rather engineering focused. But you can be an alpha user,
> > > > applying the developed method to empirical data, which deserves
> > > authorship.
> > > >
> > > > -
> > > >
> > > > What specific tasks and priorities do you currently envisage,
> > > especially
> > > > for contributors working on empirical validation and statistical
> > > evaluation?
> > > >
> > > > We will develop a Matlab application that takes .set data as inputs
> and
> > > > some figures and numbers as outputs. The core function may only take
> > > > EEG.icawinv and EEG.chanlocs.
> > > >
> > > > -
> > > >
> > > > Would validation involve datasets and analysis pipelines provided
> by
> > > the
> > > > team, public datasets, or contributors’ own data?
> > > >
> > > > Validation using empirical EEG data datasets would be nice but not
> > > > mandatory.
> > > >
> > > > -
> > > >
> > > > What technical background would you expect for these contributions,
> > and
> > > > which tools and source-localization workflows would be used?
> > > >
> > > > Knowledge of what ICA does and how it works, current dipole models
> and
> > > > various distributed source models. Experience in coding in Matlab and
> > > using
> > > > your favorite AI tools.
> > > >
> > > > -
> > > >
> > > > How would the collaboration be coordinated, and what timeline and
> > time
> > > > commitment do you anticipate? I am based in Valencia and would be
> > > > interested in collaborating remotely.
> > > >
> > > > Email and Zoom. No timeline unless Marjan has one. Location wise, I'm
> > in
> > > > Cincinnati, Ohio. Marjan is in Tehran. Cedric is in California.
> Joining
> > > > from Valencia would be nice!
> > > >
> > > > -
> > > >
> > > > What outputs are you aiming for, such as a software release or
> > > > methodological paper, and how do you envisage contributor credit
> and
> > > > authorship?
> > > >
> > > > A paper and a Matlab application. The order of the authors is Marjan,
> > > > Cedric, ...., Ilaria, Makoto, unless Marjan decides not to take the
> > > first.
> > > > My assumption is that Marjan takes the lead. We aim for NeuroImage,
> > Human
> > > > Brain Mapping, Imaging Neuroscience, European Journal of
> Neuroscience,
> > > > Journal of Neuroscience Method, Frontiers in things, etc.. If you
> know
> > > any
> > > > other good EEG methodology journals, please let me know. Maybe IEEE
> > also
> > > > works.
> > > >
> > > > The application could be transplanted to Python if someone is
> > interested
> > > in
> > > > taking the lead.
> > > >
> > > > Makoto
> > > >
> > > >
> > > >> On Tue, Sep 15, 2026 at 8:54 PM Sarvenaz Changizi <
> > > >> sarvenaz.changizi at gmail.com> wrote:
> > > >>
> > > >>
> > > >> Dear Makoto,
> > > >>
> > > >> Thank you for sharing this invitation. Your proposal to use
> anatomical
> > > ROI
> > > >> membership as a common framework across subjects and
> > source-localization
> > > >> methods sounds very interesting, and I would like to explore the
> > > >> possibility of contributing.
> > > >>
> > > >> I am a PhD candidate in psychology at the Universitat de València,
> > > Spain,
> > > >> based at ERI-Lectura. My research focuses on reading, visual word
> > > >> recognition, and psycholinguistics, and involves experimental design
> > and
> > > >> statistical modeling. I also have hands-on experience with EEG/ERP
> > data
> > > >> acquisition and have contributed to data analysis using EEGLAB in
> two
> > > >> research projects.
> > > >>
> > > >> Areas where I could contribute include empirical validation using
> > > EEG/ERP
> > > >> data, methodological input on validation design, statistical
> > evaluation
> > > and
> > > >> comparison of results across approaches, and scientific
> > interpretation,
> > > >> documentation, and manuscript preparation.
> > > >>
> > > >> Could you please share more about what you have in mind for the
> > > >> collaboration? In particular, I would appreciate clarification on
> the
> > > >> following:
> > > >>
> > > >> -
> > > >>
> > > >> What specific tasks and priorities do you currently envisage,
> > > >> especially for contributors working on empirical validation and
> > > statistical
> > > >> evaluation?
> > > >> -
> > > >>
> > > >> Would validation involve datasets and analysis pipelines provided
> by
> > > >> the team, public datasets, or contributors’ own data?
> > > >> -
> > > >>
> > > >> What technical background would you expect for these
> contributions,
> > > >> and which tools and source-localization workflows would be used?
> > > >> -
> > > >>
> > > >> How would the collaboration be coordinated, and what timeline and
> > time
> > > >> commitment do you anticipate? I am based in Valencia and would be
> > > >> interested in collaborating remotely.
> > > >> -
> > > >>
> > > >> What outputs are you aiming for, such as a software release or
> > > >> methodological paper, and how do you envisage contributor credit
> and
> > > >> authorship?
> > > >>
> > > >> I would be happy to share my CV and discuss where my background
> could
> > be
> > > >> most useful to the project.
> > > >>
> > > >>
> > > >> Thank you for your time. I look forward to hearing more about your
> > > plans.
> > > >>
> > > >>
> > > >> Best wishes,
> > > >> Sarvenaz Changizi
> > > >>
> > > >>
> > > >> PhD Candidate in Psychology
> > > >> ERI-Lectura, Universitat de València
> > > >> Valencia, Spain
> > > >>
> > > >>
> > > >>
> > > >> On 16 Sep 2026, at 00:02, Makoto Miyakoshi via eeglablist <
> > > >> eeglablist at sccn.ucsd.edu> wrote:
> > > >>
> > > >> Hi Marjan and list,
> > > >>
> > > >> I'd like to propose a small open collaboration project.
> > > >>
> > > >> The idea is that we want to use anatomical ROI membership as a
> common
> > > >> coordinate across subjects, instead of relying on conventional IC
> > > >> clustering approach in EEGLAB STUDY. In the proposed approach, users
> > > >> specify an anatomical ROI, optionally with a distance tolerance, and
> > ICs
> > > >> whose source models fall within that region are collected across
> > > subjects.
> > > >> In this implementation, anatomical ROI membership replaces IC
> > > clustering as
> > > >> the cross-subject correspondence rule.
> > > >>
> > > >> Here is a prototype I originally developed for a project with Ilaria
> > > >> Berteletti, which uses standard current dipole models. Although we
> > > >> ultimately did not use it in that study, I can tell you it worked
> > better
> > > >> than I thought.
> > > >>
> > > >>
> > > >>
> > >
> >
> https://urldefense.com/v3/__https://github.com/MakotoMiyakoshi/MMAGIC__;!!Mih3wA!AoKTTZuSu2CVqhzgmxuMpdjqVNFyIVR-LQZYBeLZwVRfqta0reT9bYiTM3uHevgACv2_Bf8HE91bDX8kg7QX9GoBaAI$
> > > >> <
> > > >>
> > >
> >
> https://urldefense.com/v3/__https://github.com/MakotoMiyakoshi/AnatomicalROIGroupICA__;!!Mih3wA!AoKTTZuSu2CVqhzgmxuMpdjqVNFyIVR-LQZYBeLZwVRfqta0reT9bYiTM3uHevgACv2_Bf8HE91bDX8kg7QXcKIZKQg$
> > > >>>
> > > >>
> > > >> I cleaned the implementation to publish here and tested it on real
> > > >> ICA/DIPFIT datasets; details are available in the repository.
> > > >>
> > > >> However, my recent work (under review) showed that more than 80% of
> > the
> > > >> pre-qualified brain ICs by IC Label were localized deeper than
> > > >> physiologically plausible cortical source depths:
> > > >>
> > > >>
> > > >>
> > >
> >
> https://urldefense.com/v3/__https://www.medrxiv.org/content/10.64898/2026.01.23.26344529v2__;!!Mih3wA!AoKTTZuSu2CVqhzgmxuMpdjqVNFyIVR-LQZYBeLZwVRfqta0reT9bYiTM3uHevgACv2_Bf8HE91bDX8kg7QXDaIe1jw$
> > > >>
> > > >> So it's time to try different cortex-constrained source estimates
> such
> > > as
> > > >> MNE, the LORETA family, LCMV, or other methods. The results can then
> > be
> > > >> compared within a common anatomical ROI space. Hence MMAGIC:
> > Multi-model
> > > >> Anatomical Grouping of Independent Components.
> > > >>
> > > >> Marjan and I would like to develop this into a
> methodological/software
> > > >> project. If anyone is interested in contributing a
> source-localization
> > > >> method, atlas-mapping approach, validation with empirical data, or
> > > >> methodological idea, you are welcome to join us. Please let us know.
> > > >>
> > > >> Makoto
> > > >> _______________________________________________
> > > >> To unsubscribe, send an empty email to
> > > >> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > > >> https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
> > > >>
> > > >>
> > > > _______________________________________________
> > > > To unsubscribe, send an empty email to
> > > eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > > https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
> > > _______________________________________________
> > > To unsubscribe, send an empty email to
> > > eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > > https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
> > _______________________________________________
> > To unsubscribe, send an empty email to
> > eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
> > _______________________________________________
> > To unsubscribe, send an empty email to
> > eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
> _______________________________________________
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
--
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