[Eeglablist] Is dipfit ever used for purposes other than localising source space data?

[Scott Makeig]

Makoto and all -

The SCALE algorithm for building head models including an estimate of skull
conductivity (the most important unknown in all electrical head models)
does NOT optimize for the *size* of the obtained IC source distributions.
Instead, it optimizes for the *compactness* of the estimated source
distributions. This is, essentially, a source distribution *shape*
parameter (penalizing distributed 'chicken pox'-like,
physiologically non-plausible source distribution estimates).

   Acar, Z.A., Acar, C.E. and Makeig, S., 2016. Simultaneous head tissue
conductivity
   and EEG source location estimation. *NeuroImage*, *124*, pp.168-180.

It should be of interest to apply this (SCALE-optimized, SCS-localized)
method to the seeming 'deep' sources we have seen in some SCCN EEG data
involving arm movements. What the results of that assessment will be I
cannot predict in advance.

Certainly, Arno is correct that dipfit estimates single (or
dual-symmetric) *equivalent
dipole* models. A single-equivalent dipole model of the scalp map of an
independent component process found by ICA decomposition is a useful
pointer to the equivalent of the 'electrical center of gravity' of the
active patch of cortex from which the source activity is projecting.
Michael Scherg clearly explained all this in his 1990 chapter

   Scherg, M., 1990. Fundamentals of dipole source potential analysis.
*   Auditory evoked magnetic fields and electric potentials. Advances in
audiology*, *6*, pp.40-69.

although its physical usefulness for EEG source localization was first
pointed out by physicists in the 1960s, as I recall.

As a simple equivalent example. one might define an 'equivalent dipole' for
your current position on earth (using GPS coordinates, say) with its
orientation pointing out from the center of your head through the bridge of
your nose.  This could well model your position and (head) orientation in
3D space, though It would *not* capture any information about your body
*size* (height, weight) or shape.

One can attempt to fit a single-equivalent dipole model to any EEG scalp
map, although very few maps are 'dipolar', meaning they match the
projection of a brain-centered single equivalent dipole model. In the past,
EEG researchers attempted to fit scalp maps representing peaks in
trial-average ERPs using equivalent dipoles. Typically, the fits were poor
- this is because the evoking events perturbed potential dynamics in
several or many parts of cortex in different ways, so the average ERP peak
maps represented combinations of source projections - and, unfortunately,
fitting multiple equivalent dipoles to such maps is under-determined,
meaning many different spatial muli-dipole solutions could account for the
same maps equally well.

ICA decomposition avoids this problem by finding component processes with
fixed scalp projection patterns that account for some portion of the entire
scalp data, whose activity time courses are distinct from all the remaining
EEG.  As Arno pointed out, many of the brain-based IC source process scalp
projection maps found by adequate ICA decomposition of sufficient data are
almost perfectly 'dipolar' (i.e., exhibiting a difference between the
best-fitting brain single-equivalent dipole model and the data-driven IC
scalp map of as low as ~1%). This is compatible with a conclusion that the
activity accounted for by the IC arises, entirely or nearso, in spatially
near-synchronous activity across *some* cortical source patch, whose
spatial *center* and central *orientation* is estimated by the equivalent
dipole position and orientation. This model is supported by the fact that
cortical connectivity is highly weighted toward short-range,
neighbor-to-neighbor connections, allowing the emergence of areal synchrony
or near-sychrony in local field potential.

The SCS (Sparse, Compact, Smooth) algorithm that Zeynep Akalin Acar and I
use (in the NFT plug-in) for distributed source estimation also takes into
account the physical orientations of each cortical surface model voxel,
which strongly limits the physiologically possible shapes and sizes of the
source patch estimates -- more so as the electrical current flow head model
used in the analysis is more accurate.

   Acar, Z.A. and Makeig, S., 2013. Effects of forward model errors on EEG
source localization.
*   Brain topography*, *26*(3), pp.378-396.

Scott

On Mon, Jul 22, 2019 at 7:34 PM Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
wrote:

> Dear Arno,
>
> > We have never claimed that ICA components localize to point like regions
> (the flux of current in a small patch of cortex can be well represented
> with a point like dipole at a slightly deeper location
>
> SCCN has been claiming what I call 'small patch hypothesis' for the
> ICA-resolved cortical effective source. For example, see Delorme et al.
> (2012).
>
> ...There are strong biological reasons to believe that under favorable
> circumstances ICA should separate signals arising from local field
> activities in physically distinct, compact cortical source areas: First,
> short-range (<100 micrometer) lateral connections between cortical neurons
> are vastly more dense than longer-range connections [3,4]...
>
> The extremity of this view is SCALE algorithm by Akalin-Acar and Makeig
> where patch size serves as a cost function to be minimized by changing
> skull conductivity as independent variable.
>
> According to Nunez and Srinivasan (2006), cortical patch size should be > 1
> inch^2 to be detected at the scalp sensor. See my talk on
> https://urldefense.proofpoint.com/v2/url?u=https-3A__youtu.be_lRyZxt2WeKk-3Ft-3D1392&d=DwIFaQ&c=-35OiAkTchMrZOngvJPOeA&r=kB5f6DjXkuOQpM1bq5OFA9kKiQyNm1p6x6e36h3EglE&m=5-eyj2sXmrSMMy-Yt23PTby77cd_YWeIO_T7kslU9eoP_voMS7d8LCuVBoCUhXQR&s=LCH7tXRj2g1BNNgigC4tNPDOyjJVmcsIxD4JPRBXmgk&e=  Also Nunez and Srinivasan (2006) Fig.
> 1-20.
>
> Makoto
>
>
>
> On Fri, Jul 12, 2019 at 9:12 PM Arnaud Delorme <arno at ucsd.edu> wrote:
>
> > > Let me tell you dirty secret of dipfit.
> > > In order for cortical activation to be recorded at the scalp level
> > without
> > > averaging, the patch size should be > 1 inch^2 (Nunez and Srinivasan,
> > > 2006). However, dipole model assumes a spatial point with 6 parameters.
> > > Therefore, estimating dipolar sources of scalp-recorded potential means
> > to
> > > model spatially spread cortical patch with a point. This is a
> fundamental
> > > violation of the dipole model.
> >
> > The dipole model is a mathematical tool that works well to
> > summarize/compress data (6 values; 3 position; 2 moment; 1 amplitude) can
> > often explain scalp topographies with 256 channels with 99% accuracy
> which
> > indicates that this is a “good” model (6 numbers explaining 256 numbers).
> > We have never claimed that ICA components localize to point like regions
> > (the flux of current in a small patch of cortex can be well represented
> > with a point like dipole at a slightly deeper location - I believe it is
> > mathematically equivalent but would welcome validation from a source
> > localization expert).
> >
> > If you do not like dipoles, in Dipfit 3 you can also use eLoreta to
> > calculate distributed source models associated with ICA components.
> >
> > Best wishes,
> >
> > Arno
> >
> > _______________________________________________
> > 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
>
>
>
> --
> Makoto Miyakoshi
> Assistant Project Scientist, Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
> _______________________________________________
> 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