[Eeglablist] Using ICA with interpolated channels

Ronald Phlypo Ronald.Phlypo at ugent.be
Fri Oct 1 01:16:00 PDT 2010


  Dear Jordi and list members,


Personally, I do not fully agree that one does not benefit from spatial 
interpolation of the EEG before ICA analysis. Whilst the comment

<<Hence, a new dimension of data is not being created by doing the
interpolation.>>

does certainly hold true for linear interpolation schemes, this does no 
longer hold for nonlinear interpolation schemes (cubic splines, 
wavelets, etc.), since a linear unmixing process would see new data in 
there. At best, the dimension of the "cerebral data" would not increase, 
but it might well be that the signal to noise ratio augments by 
increasing the number of virtual channels (obtained through 
interpolation) included in your dataset, where I made the implicit 
assumption that the interpolation is a good approximation to the 
potential field, i.e., that the potential field has been well sampled 
from the start by choosing good electrode locations. The total dimension 
of your data thus increases by introducing new noise sources (which are 
the interpolation errors).

A question that arises to me is whether interpolation should be used or 
best least squares fitting (since the electrode measurements themselves 
include noise, which should not to be modelled by the "interpolated" 
potential field).

Hope this keeps the discussion alive and I would very much be interested 
if anyone would already have obtained some results on this !


Ronald




Le 28/09/2010 18:03, Philip Michael Zeman a écrit :
> Hello Jordi
>
> there are a number of methods of interpolation and not all are equal.
> However, generally:
>
> an interpolated channel is the linear combination of 2 or more other real
> EEG channels.
>
> Hence, a new dimension of data is not being created by doing the
> interpolation.
>
> Hence, the interpolated channel does not benefit the ICA process.
>
> This said,
>
> it is pemissible to interpolate channels of back-projected (scalp projected)
> components "after" applying ICA.
>
>
> Phil
>
>
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> Philip Michael Zeman B.Eng, Ph.D.
> Applied Brain and Vision Sciences Inc.
> Brain Function Analysis for Novel Paradigms and Serious Games
> Analysis of Pharmaceutical Effects on Brain Function
> http://www.abvsciences.com
> Latest Brain Research Result:
> http://www.spatialbrain.com
> Email:   pzeman at alumni.uvic.ca
> Phone: +1-250-589-4234
> LinkedIn Profile: http://ca.linkedin.com/in/philipmichaelzeman
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>
> ----- Original Message -----
> From: "Jordi Costa Faidella"<jordicostafa at gmail.com>
> To:<eeglablist at sccn.ucsd.edu>
> Sent: Wednesday, September 22, 2010 5:43 PM
> Subject: [Eeglablist] Using ICA with interpolated channels
>
>
>> Dear EEGlab users,
>>
>> this question arose when reading the new FASTER method, but I think is of
>> general importance for all of us. Is it correct to perform an ICA on a
>> dataset in which some of the channels have been interpolated?
>>
>> thanks,
>>
>> Jordi
>> _______________________________________________
>> 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
> _______________________________________________
> 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


More information about the eeglablist mailing list