[Eeglablist] Source localization and hippocanpus

[Cedric Cannard]

Hi Makoto,

Exciting development. Thanks for working on this, I've always thought it is a big gap in the EEG field to have such questions unanswered.

Quick curiosity questions:
1) why not including subiculum?
2) is this only using dipole fitting tools? what about other methods like eLoreta or LCMV beamforming?
3) "Destrieux Atlas is based on Colin27, which is a single-subject head model and has annotated gyri and sulci. But it does not come with hippocampi. So I used HippUnfold (DeKraker et al., 2022, 2023) to auto-demarcate hippocampi on the Destrieux Atlas."

-> I'm surprised, I remember clearly the Colin atlas having the hippocampi included when using the ROIconnect plugin for LCMV beamforming, which I had to exclude in the code as I suspected it was not reliable, but it was there.

-> is a single subject head model the way to go for validating this kind of thing? aren't the findings vulnerable to be specific to this person only?


Cedric


Sent with Proton Mail secure email.

On Friday, August 21st, 2026 at 1:00 PM, Cedric Cannard <ccannard at protonmail.com> wrote:

> Hi Makoto,
> 
> Exciting development. Thanks for working on this, I've always thought it is a big gap in the EEG field to have such questions unanswered.
> 
> Quick curiosity questions:
> 1) why not including subiculum?
> 2) is this only using dipole fitting tools? what about other methods like eLoreta or LCMV beamforming?
> 3) "Destrieux Atlas is based on Colin27, which is a single-subject head model and has annotated gyri and sulci. But it does not come with hippocampi. So I used HippUnfold (DeKraker et al., 2022, 2023) to auto-demarcate hippocampi on the Destrieux Atlas."
> 
> -> I'm surprised, I remember clearly the Colin atlas having the hippocampi included when using the ROIconnect plugin for LCMV beamforming, which I had to exclude in the code as I suspected it was not reliable, but it was there.
> 
> -> Are you sure using an automated method like HippUnfold is fully reliable?
> 
> -> is a single subject head model the way to go for validating this kind of thing? aren't the findings vulnerable to be specific to this person only?
> 
> 
> Cedric
> 
> 
> Sent with Proton Mail secure email.
> 
> On Friday, August 21st, 2026 at 10:48 AM, Евгений Машеров via eeglablist <eeglablist at sccn.ucsd.edu> wrote:
> 
> > First of all, thank you very much for this fascinating and important work.
> > The threshold for EEG amplitude that allows signal registration, 1.65 μV, seems reasonable. By comparison, when diagnosing brain death, an amplitude of oscillations of 2-3 μV does not prove brain function and is considered an artifact.
> > However, the conclusion regarding whether signals from the hippocampus can be recorded on the scalp relies on several assumptions, and these can be challenged. The first is that the source is necessarily a dipole. The second is that it is a single dipole, not a set of synchronized dipoles distributed over a certain area and similarly oriented.
> > The question of the validity of these assumptions arose in my mind quite a while ago, when I was tasked with using the BrainLoc dipole localization program (developed by Yuri Koptelov) to analyze clinical EEGs. I was immediately skeptical about the possibility of recording a signal from a non-cortical dipole on the scalp, since the dipole potential decreases inversely with the square of the distance. However, practical experience has shown that sources deep in the brain are localized quite accurately, and their location is confirmed by recordings during surgery or from implanted electrodes.
> > In an attempt to explain this paradox, I proposed two explanations.
> > 1. In addition to the dipole component, a monopole component also contributes to the EEG. Its potential decreases less rapidly with distance and can therefore be recorded at a greater distance. A possible source of this component could be fluctuations in the ionic composition inside and outside the neuron (the contribution of glia may also be significant).
> > https://urldefense.com/v3/__https://link.springer.com/article/10.1134/S0006350919030138__;!!Mih3wA!GQ6tkO0-y-UdiBpSSa5dvJ5WwALJ32XNjpXJS-5N7m99k4m-659LlesRaG8YmhUmYS88D71nxC6knyaGTmLBp4yz1xU$
> >
> > 2. The potential from a system of dipoles changes differently than from a single dipole and propagates over a greater distance. Likewise, the potential from a system of quadrupoles can propagate over a greater distance than from a single dipole.
> > https://urldefense.com/v3/__https://link.springer.com/article/10.1134/S0006350921040114__;!!Mih3wA!GQ6tkO0-y-UdiBpSSa5dvJ5WwALJ32XNjpXJS-5N7m99k4m-659LlesRaG8YmhUmYS88D71nxC6knyaGTmLBvqm70Fw$
> >
> > It is quite possible that both of these mechanisms need to be taken into account, and even then the complexity of the EEG generation process may not be exhausted.
> >
> > Your truly
> >
> > Eugen Masherov
> >
> > > Hi list,
> > >
> > > Earlier this year, we discussed whether scalp EEG can measure
> > > hippocampus. Komal
> > > and I developed a forward model simulator, L-EEGiability Atlas, to answer
> > > this question. The Github repository is still private. It'll be published
> > > after Komal registers it to Zenodo. This is a follow up. Technical detail
> > > of this forward model is provided after my signature.
> > >
> > > In this forward model simulation, the parameters uses are:
> > >
> > > Dipole moment: 0.17 nAm/mm^2 (which is the strength of sleep spindle)
> > > Uniform recruitment rate: 66.6% (arbitrary choice, but for a specific
> > > reason--you will find the reason below)
> > > Selected regions: CA1, CA2, CA3, CA4 in the left hemisphere (no subiculum)
> > > Total ROI area: 9.59 cm^2
> > >
> > > A critical fact is that EEG signals smaller than 1.65 microV are not
> > > guaranteed to be measured (Nuwer et al., 1998). I guess this reflects
> > > realistic requirements of an EEG recorder's noise floor. Thus, < 1.65
> > > microV means 'not eligible for scalp EEG'.
> > >
> > > In this condition, the maximum scalp potential observed is 1.649 microV!
> > >
> > > Thus, to measure hippocampal EEG, average dipole moment needs to be > 0.17
> > > nAm/mm^2 and/or average recruitment rate needs to be > 66.6%. Our next
> > > question is therefore: can we obtain empirical evidence supporting these
> > > numbers?
> > >
> > > By the way, if the left hippocampus becomes epileptic (uniform recruitment
> > > 100%, 0.77 nAm/mm^2), you would measure 11.126 microV at the scalp. If I
> > > apply the Murakami-Okada limit of 2 nAm/mm^2, the measured signal would be
> > > 19.402 microV.
> > >
> > > Here is the comparison table to get a feeling of how much depth and folding
> > > penalty the hippocampus EEG undergoes
> > > (all left hemisphere ROIs, 0.17 nAm/mm^2, uniform 66.6% recruitment,
> > > amplitude shown is positive peak only at the scalp maximal, three-layer
> > > BSCR=40 model)
> > >
> > > Hippocampus, 9.59 cm^2, 1.649 microV
> > >
> > > Sup Occipital Gyrus, 15.71 cm^2, 36.283 microV
> > > Sup Parietal Gyrus, 34.94 cm^2, 42.330 microV
> > > Postcentral Gyrus, 28.94 cm^2, 26.263 microV
> > > Sup Temporal Gyrus, 24,17 cm^2, 24.420 microV
> > > Inf Front Opercular, 19.75 cm^2, 18.194 microV
> > >
> > > Insula, 30.77 cm^2, 16.238 microV
> > >
> > > All pure Gyri, 784.19 cm^2, 49.404 microV
> > > All pure Sulci, 338.03 cm^2, 48.325 microV
> > >
> > > Whole brain, 2348.56 cm^2, 29.956 microV
> > >
> > > Angular + Supramarginal + Sup Parietal + IntraParietal Sulcus, 113.42 cm^2,
> > > 80.957 microV
> > >
> > > Note that the area of the selected ROI and the scalp-measured EEG amplitude
> > > do not correlate. This is because (1) cancellations due to foldings, (2)
> > > the effect of volume conduction is actually limited locally so that source
> > > effects do not add up at one measurement point.
> > >
> > > If you have a request to test some hypothesis/question, let me know.
> > > L-EEGiability will be publically available in a week or two.
> > >
> > > Makoto
> > >
> > > %%%%%%%%%%%%%%%%%%%%%%
> > > Technical note
> > >
> > > Leadfield-based EEG eligibility (L-EEGiability, pronounced as
> > > 'eligibility') Atlas is an extended version of the Destrieux Atlas
> > > (Destrieux et al., 2011). Destrieux Atlas is based on Colin27, which is a
> > > single-subject head model and has annotated gyri and sulci. But it does not
> > > come with hippocampi. So I used HippUnfold (DeKraker et al., 2022, 2023) to
> > > auto-demarcate hippocampi on the Destrieux Atlas.
> > >
> > > Then I determined a physiological plausible range of current dipole moments
> > > in nAm/mm^2 for the human brain (Murakami & Okada, 2015; Inverso et al.,
> > > 2016; Rosen et al., 2019). The reported values are like 0.17 for sleep
> > > spindle, 0.2-0.35 for visual evoked potential at V1, and 0.16-0.77 for
> > > interictal spikes. Murakami & Okada also proposed a physiological limit of
> > > 1-2 nAm/mm^2.
> > >
> > > I also set a few other open parameters, such as synchronization rate. This
> > > is completely arbitrary. We need empirical data to fill this in.
> > >
> > > The brain-to-skull-conductivity (BSCR) ratio precalculated are 26, 40, 80.
> > > Although BSCR=26 is the modern estimate, because the model is three tissue
> > > layers without CSF, using BSCR=40 compensates for the lack of CSF layer
> > > (EFB Chapter 6).
> > >
> > > On Fri, Jan 30, 2026 at 5:29 PM Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
> > > wrote:
> > >
> > >> Hi all,
> > >>
> > >> I've read this discussion with great interest!
> > >> Here are my thoughts.
> > >>
> > >> 1. If I adopt a conventional dogma of generative mechanism of EEG
> > >> i.e., all/most of scalp-measured EEG signals entirely generated by
> > >> post-synaptic membrane potential in cortical surface (Electric Fields of
> > >> the Brain, EFB, by Nunez and Srinivasan 2006 adopts this assumption, for
> > >> example), measuring EEG signals generated in the hippocampus using a
> > >> conventional EEG recorder (i.e., sensitivity limit > 1 microV) would be
> > >> impossible primarily because of the geometry (too small, too deep, rolled
> > >> shape, etc..)
> > >> 2. However, a modern electrodiffusive neural-extracellular-glia
> > >> (edNEG) model indicates the possibility that non-synaptic source activity
> > >> affects scalp EEG as well, which is much less studied. See Saetra et al.
> > >> (2021) for full details (but this one is super technical).
> > >> https://urldefense.com/v3/__https://journals.plos.org/ploscompbiol/article?id=10.1371*journal.pcbi.1008143*libraryItemId=11397962__;LyM!!Mih3wA!BJz_jI7amVJIO4yLx16kUNcvtKfKz2Vq2a8fn-buH6yEmuldbNRSNq8JZ679B42teXAZ4sz3nMIKlNimhKtMlKPZmVs$
> > >> 3. I read the dispute between Joseph's group and Mike's group with
> > >> great interest. If I stand on the conventional viewpoint, I agree with
> > >> Mike, and I usually stand on it. But that does not mean non-conventional
> > >> source cannot exist, particularly if it is observed in a low-frequency
> > >> range (below delta). After all, we do not know EEG phenomenon completely,
> > >> so we are not there yet to make an a priori prediction that can properly
> > >> bound the observation (i.e., "If we observe X under certain conditions, it
> > >> MUST BE noise because it cannot be there from biophysics!")
> > >> 4. That said, I still want to emphasize that learning the conventional
> > >> synaptic dogma and EEG's unique scaling law (i.e., volume conductor theory,
> > >> including what I call 'transducer array effect' of a dipole layer) is very
> > >> important. For those who are curious, below I provide two sources of
> > >> information:
> > >>
> > >> The first material to read is an excerpt from EFB 2nd ed page 81-83. Here,
> > >> the authors demonstrate a typical misconception/misuse of both equivalent
> > >> current dipole model and assumption of EEG source distributions (i.e.,
> > >> small vs. large cortical patches). As a result, such an incorrect
> > >> application leads to an implicit conclusion that there was a 5V (=5,000,000
> > >> microV) source inside the brain. The authors' following concluding remark
> > >> suggests there exists this known pitfall/misconception in literature
> > >> reporting dipole source models in general (ouch...)
> > >>
> > >> *Of course, such extracellular potentials are too large by a factor of
> > >> 1000 or more to be physiologically realistic. However, such considerations
> > >> have not prevented reports of such magical dipoles. One wonders if this
> > >> explains why source magnitudes are often not reported.*
> > >>
> > >> https://urldefense.com/v3/__https://drive.google.com/open?id=1LGyydB9ZucbBG9EQoZDKAg4CEUJlAWon__;!!Mih3wA!BJz_jI7amVJIO4yLx16kUNcvtKfKz2Vq2a8fn-buH6yEmuldbNRSNq8JZ679B42teXAZ4sz3nMIKlNimhKtMHlzFVrk$
> > >>
> > >> The second material to read is my manuscript under review, in which I
> > >> described what I believe is the correct 'scaling law' of scalp-recorded EEG
> > >> signals. Please find the section "Where did the small patch model come
> > >> from, and what does it miss?" in Discussion.
> > >>
> > >> https://urldefense.com/v3/__https://www.medrxiv.org/content/10.64898/2026.01.23.26344529v1*libraryItemId=18612094__;Iw!!Mih3wA!BJz_jI7amVJIO4yLx16kUNcvtKfKz2Vq2a8fn-buH6yEmuldbNRSNq8JZ679B42teXAZ4sz3nMIKlNimhKtMsEnkBo8$
> > >>
> > >> I will submit a separate post about this manuscript, as this 'scaling law'
> > >> is a tool for a larger aim.
> > >>
> > >> I have one more thing I wanted to report here, which would make a
> > >> practical (and I believe substantial) contribution to this debate. But
> > >> because Komal wants to keep it a low profile until he comes up with some
> > >> proof of concept, I cannot tell you what it is ha ha. I believe that this
> > >> solution provides a clear answer to the question "Can EEG measure the
> > >> hippocampus' at least within the limitation of the conventional synaptic
> > >> dogma + volume conductor theory.
> > >>
> > >> 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 .