[Eeglablist] Can EEGLAB skip ICA decomposition prior to source localization?

[Scott Makeig]

Min -

EEGLB DIPFIT functions (called from the commandline) will work on any scalp
map, but will only attempt to find a single-dipole model (or else, a
bilaterally symmetric two-dipole model). Fitting more dipoles can be done
using the underlying functions, but is inherently subject to multiple (local
minima) solutions.

To find a scalp map that represents the output of a single source area (with
a single equivalent dipole model) is not easy. Often, researchers hope that
this situation may obtain at peak excursions in average ERP waveforms.
However, very soon after (if not coincident with) first arrival of sensory
information in cortex,  statistics of EEG activity in many parts of the
brain are affected, meaning later portions of ERPs may typically sum
activity arising in multiple cortical areas. This leaves the problem of
finding out which areas are participating in the ERP, and with which time
courses. This problem is technically difficult, since it is so often
spatiotemporally complex.

ICA isolates temporally independent sources in the data. Because of cortical
connectivity patterns, it is far most likely that synchronized (or partially
synchronized) field activity within a cortical patch produces a (far-field)
scalp signal (on most of the electrodes) that is maximally independent of
similar activities produced in other cortical patches (or in non-brain
sources such as scalp muscles). In fact, and compatible with this
physiological model, ICA component maps typically are well matched to an
equivalent single-dipole model. Use the function envtopo()  (under the menu,
Plot > Component ERP with scalp maps) to see how ICA decomposes a given ERP
(of course, after running ICA on the data -- doing this correctly this takes
some care; study the tutorial and relevant papers).

I and co-authors have discussed these things at more length in a TICS review
article:
http://sccn.ucsd.edu/~scott/pdf/TICS04_Preprint.pdf<http://sccn.ucsd.edu/%7Escott/pdf/TICS04_Preprint.pdf>
(0.5MB)

Practical details are discussed in a book chapter:
http://sccn.ucsd.edu/%7Escott/pdf/PBR06.pdf      (36 MB)

If you are interested in highest-quality dipole fits (now) and cortical
source estimates (in future, or using other existing software), then look on
the EEGLAB home page for a poster previewing a forward head modeling toolbox
that Zeynep Akalin Acar at SCCN is about to release.

Best luck,
Scott Makeig


On Tue, Dec 30, 2008 at 4:12 PM, Min Bao <ustcbm at yahoo.com.cn> wrote:
Hi experts,

I am new to source localization analysis. I've read the eeglab tutorial on
dipfit. It seems ICA is necessary before doing dipfit. However, the ICA
usually generates N components if there are N electrodes in the data.
Selecting the appropriate component for dipfit is such a subjective step
which makes a newbie like me very confused and uncomfortable. Could anyone
tell me why the ICA is a necessary step prior to dipfit? Can we do source
localization merely on the averaged epoch (ERP) rather than an independent
component? Thanks a lot for any direction!

Best wishes,
Min Bao


Scott Makeig, Research Scientist and Director, Swartz Center for
Computational Neuroscience, Institute for Neural Computation, University of
California San Diego, La Jolla CA 92093-0961, http://sccn.ucsd.edu/~scott
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20081231/df2bf1cf/attachment.html>