[Eeglablist] bad channels removal before ICA

[Arnaud Delorme]

You should not use interpolated data channel for ICA (since it will 
reduce the dimentionality of the data).
If you absolutely want to interpolate channel activity at specified 
electrodes (the graphical interface of EEGLAB cannot do it but the 
topoplot function can interpolate data at any location (2-D 
interpolation, not spline)), you should use the PCA option of runica to 
reduce the number of dimension of the data prior to running ICA (for 
instance, if you interpolate 2 channels in your 64-channel data, enter " 
'pca', 62 " in the edit box for runica option).

Arno

Darren Weber wrote:

>It is acceptable for avg ERP to replace bad channels with the _weighted_
>average from surrounding electrodes (weighted by distance, either linear or
>arc length distance, so it's best if you have electrode coordinates for this
>calculation).  A better method may be to interpolate the timeseries at the
>bad channel from a surface spline of all good channels.  I recall reading
>this in a statement paper on methods (can't recall the reference just now).
>Obvioulsy it is better to have good measured data, but we don't always live
>in that world ;-)  I'm not sure what the implications of this might be for
>ICA, my comments are just based on reading of acceptable methods in avg
>ERPs.
>
>Best, Darren
>
>  
>
**

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20031105/bf5b17b5/attachment.htm