[Eeglablist] ICA of concatenated subjects?

[Guillaume Lio]

Hi Ana,

A mandatory assumption for blind source separation (BSS) algorithms like
ICA is that the mixing matrix remains invariant, i.e., that the sources,
electrodes and geometry of the head do not change during the experiment.

This issue is critical with the group ICA (gICA) approach. By concatening
all the subject's data, the 'mixing matrix' changes for each subject
because of interindividual anatomo-functional variability. That's why you
can have poor results with some class of BSS algorithm, even with very
clean data.

A potential solution is to use dimension reduction techniques like PCA, to
reduce the differences between subjects. (e.g. Eichele et al. , GIFT and
EEGIFT toolboxes, ...).

An other solution is to avoid the dimension reduction step with BSS/ICA
algorithms naturally robust to mixing matrix distortions, like
second-order statistics based algorithms (SOBI, UWSOBI...).

You can find more explanations and demonstrations in our recent study :

Lio G, Boulinguez P.
Greater robustness of second order statistics than higher order statistics
algorithms to distortions of the mixing matrix in blind source separation
of human EEG: implications for single-subject and group analyses.
Neuroimage. 2013 Feb 15;67:137-52.

If you have any questions about the paper and/or the algorithms used,
please ask.

Hope this helps,

Guillaume.





> Hi,
> I was wondering if there is any problem with trying ICA for the entire
> group of subjects (as it's usually done in fMRI) instead of running ICA
> for
> each individual subject and then finding the common components with a
> cluster analysis. I thought it could be as simple as concatenating all the
> subject's data, but I tried it and the ICA components that I got from it
> are looking quite bad. They look as if the data were really noisy. I
> understand that there is more variability, but this data is very clean and
> I expected something better. Am I missing something? I appreciate any
> help.
> Many thanks,
> Ana
> _______________________________________________
> 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