[Eeglablist] electrode mislocalization

[Robert Oostenveld]

Hi Rodrigo,

On 22 Nov 2006, at 11:02, Rodrigo Segnini wrote:
> How important is to have a very precise electrode localization map,  
> the
> like you would produce for every subject using, say, the Polhemus  
> system?

I would say that it is important, but it depends on your goals.

> The following article states, for example, an effect on source
> localisation, but that is without using ICA.
> ...

Another study showing it is
  Khosla D, Don M, Kwong B.
  Spatial mislocalization of EEG electrodes -- effects on accuracy of  
dipole estimation.
  Clin Neurophysiol. 1999 Feb;110(2):261-71.

The statements in both papers also apply to ICA, except that in ICA  
it is more difficult to classify signal to noise ratio (since that  
depends on the quality of the ICA unmixing).

> You could state your improvement in percentual terms.

Dipole location is not expressed in percent, but in mm or cm. The  
accuracy of source localization depends on multiple factors:
- noise in the data
- location of the source (deep vs. superficial)
- accuracy of the source model (i.e. can the active tissue be  
represented as an infinitely small point source)
- accuracy of the volume conduction model geometry (sphere, vs.  
standard-realistic, vs. individual realistic)
- accuracy of the volume conduction model tissue conductivities
- accuracy of the volume conduction model numerical computation
- accuracy of the location of the electrodes in the model
- size of the recording area (i.e. whether electrodes are point-like,  
or whether they average over larger pieces of scalp)
- whether you interpret the source location on a template anatomical  
MRI or on the individuals MRI
- accuracy of the coregistration of the forward (source+volume) model  
with anatomical MRI
The list above probably is not even complete yet.

What I would be mainly concerned about is the volume conduction model  
errors.

Furthermore, I would also be concerned about systematic  
mislocalization of electrodes: Imagine the following: If you apply a  
cap that is much too small for the subject, the electrodes all will  
be higher on the head (since you cannot pull the cap sufficiently  
downward). That means that the scalp distribution of a known source  
will project _stronger_ on the lower electrodes on the cap (since  
those electrodes are too high on the head). That means if you do not  
take the placement of the electrodes into acocunt but use standard  
locations, the estimated source location will be deeper than the true  
location.

Hence systematic errors are the most problematic. Even if you do all  
source modelling to the best capabilities using standard software and  
only have "random" errors in your data and electrode locations, I  
would never trust the source location of EEG data to be more than  
~2cm accurate. That is not to say that it can be done more accurate,  
but that requires considerably more effort than I expect most EEGLAB/ 
DIPFIT users to put into source localization. Remember that the  
strong point of EEG is its temporal resolution: the source timecourse  
is what distinguishes EEG from fMRI, and timecourse separation is the  
strong point of ICA.

best regards,
Robert