[Eeglablist] Using ICA with interpolated channels

[Hugh Nolan]

Hi Jordi,
In response to your question "Is this PCA that constrains the possible ICs
implemented by default in the method or it is something you had to perform
in order to be able to use FASTER with your dataset?", the PCA reduction is
performed by default in FASTER when ICA is run, to constrain the number of
independent components computed based on a) the k-value, a measure of the
number of points per IC (as mentioned in the EEGLAB handbook, and
"Information-based
modeling of event-related brain dynamics." by Onton et al), and b) the
number of interpolated channels, as mentioned previously. Is this necessary
to be able to use FASTER? Well, during testing of the FASTER method we ran
tests of various k-values for a number of datasets, using 15, 25, 40 and
also 1, to see the effect on the output. Generally, the high variance
components, particularly the EOG component, was little affected by this, but
the lower variance components were - this included pop-offs and other
transient artifacts. These lower variance components were often split into
numerous components, which made detection of artifacts difficult. It's a
difficult issue, to determine the correct number of sources - we found that
the value of k = 25 gave good results, so we stuck with that. Of course,
experimentation could show a different value works better on your own
particular dataset - if so, let us know, we are always eager to hear.
Hugh and Rob

-- 
Hugh Nolan,
Trinity Centre for Bioengineering,
Printing House,
Trinity College Dublin.

Tel: +353861297722
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20100929/8939e691/attachment.html>