[Eeglablist] Interpolation and ICA

Bernhard Meyer a0302186 at unet.univie.ac.at
Fri Jan 8 07:55:01 PST 2010


Dear All,

I appreciate the information I got and want to conclude the common 
recommendation to calculate ICA, remove artifactual ICA components and 
then interpolate channels. Furthermore I want to add following thoughts:

1.) The broken or unplugged channel should be removed before 
ICA-calculation, as Andreas Widmann wrote. Since I combined removing the 
channel and interpolation, it was important for me to have that explicit.

2.) Consequently using the interpolation in pop_precomp() is the last 
chance to remove artifactual-clusters. Maybe it would be worth to add 
that information.

3.) If ICA-components are not interpolated, there is no need for 
interpolation in ICA-exclusive research. E.g. if I don't plot ERPs of 
that channel but I want to identify a component.

4.) To interpolate a part of a bad channel (e.g. during one block of 
epochs with no signal from broken electrode) does not seem to be 
possible up to now. What about the idea (Point b Andreas Widmann) to 
fill that part of the channel with noise or to just remove it before 
ICA-calculation and interpolate that segment after?

It would be kind if you evaluate these points to prevent possible 
misconceptions.

Best,
Bernhard



More information about the eeglablist mailing list