[Eeglablist] question: determining number of generating sources

[Arnaud Delorme]

Dear Jon,

A better way to estimate the number of source using eigenvalues of the 
autocorrelation matrix may be found in

   WACKERMANN, J. 1996. Beyond mapping: estimating complexity
   of multichannel EEG recordings. Acta Neurobiologiae
   Experimentalis, 56, 197-208.
   WACKERMANN, J. 1999. Towards a quantitative characterization
   of functional states of the brain: from non-linear methodology
   to the global linear description. International Journal of
   Psychophysiology, 34, 65-80.

This is implemented in the numdim() function of EEGLAB. In general 
though, it is better to use as many source as possible. In theory every 
single neuron is an independent source. The contribution of sources 
ordered by their amplitude would most likely resemble an exponentially 
decreasing function. ICA just imposes a cut-off on the number of sources.

The conclusion of the paper you attached lacks experimental testing. It 
is based on the assumption that there is a finite number of EEG source 
in the brain which is not the most common belief (see above).

The fact that ICA cannot separate EEG sources and artifacts because the 
EEG signal does not have a finite number of stationary sources is simply 
not true because it does. It is true that this area of research would 
require more investigation, but this article does not bring any 
conclusive argument. It is misleading at the most.

Arno

halfordj at musc.edu wrote:
> Dear EEGLabList,
>
> In a recently published paper on ICA of clinical EEG (Unsworth 2006 J of Clin
> Neurophys; attached to this email), the authors discuss a method for
> determining the number of source signals in a blind signal source separation
> window by using eigenvalue analysis. Would this be the same as looking at a
> plot of the %variances of the ICs produced by an ICA analysis?
>
> Thanks,
> Jon
>