[Eeglablist] Interpolating bad channels while avoiding PCA before ICA

[Andreas Widmann]

Hi Jeff and list,

> Concerning channel interpolation, I have wondered about using non-linear spatial interpolation, specifically if one can avoid reducing rank by using it.
The interpolation of a channel itself does not reduce the rank, but it also does not increase it. The rank is reduced by the loss of the information encoded in the bad channel. As far as I understand it, it cannot be retrieved by any interpolation. Please correct, if this statement is wrong.

> We all know that linear interpolation is bad in this regard, but the better nonlinear interpolation methods, based on mathematical or physical theory, may provide a way around this. I am thinking of spherical splines, spherical harmonics, or 3D splines. Has anyone out there tried this? I assume most commercial and academic interpolation use linear interpolation, so one would have to make a special effort.
Spherical spline interpolation is essentially a linear operation (the name is somewhat misleading; e.g., when interpolating a channel, the sum of the interpolations of a number of component back-projections is the same as the interpolation of the sum of a number of back-projections).

Best,
Andreas

> -Jeff Eriksen
> OHSU
> 
> From: Mikołaj Magnuski <imponderabilion at gmail.com>
> Date: Sunday, May 19, 2013 4:11 PM
> To: "eeglablist at sccn.ucsd.edu" <eeglablist at sccn.ucsd.edu>
> Subject: [Eeglablist] Interpolating bad channels while avoiding PCA before	ICA
> 
> Dear EEGLab community,
> 
> Given that PCA is not recommended before ICA,
> how to avoid it while still interpolating bad channels?
> 
> My initial analysis of the data is in channel space
> so I would like to have all channels for all participants.
> Therefore I want to interpolate bad channels (I have a
> maximum of 4 bad channels out of 64 per subject).
> This causes the rank of the data to decrease and compli-
> cates using ICA for artifactual components removal.
> 
> Do you think a following way of avoiding PCA before ICA is ok?
> 1. preprocess the data before ICA
> 2. perform ICA and remove artifactual components
> 3. interpolate channels
> 
> So, in short: is it valid to interpolate channels only after
> removal of artifactual ICs?
> 
> Mikolaj M
> _______________________________________________
> 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