[Eeglablist] Topographic plot question

[Arnaud Delorme]

Dear Thanasis,

> I'm interested in plotting the topographic map of a set of specific 
> independent components that ICA has found on some EEG recordings. The 
> number of EEG electrodes is 64. However the problem I face, is that 
> the only data that I have, is a subset of N independent components 
> where N<number of electrodes together with the N vectors of the ICA 
> weights (W has dimension of Nx64) and the channel locations.
> Would it be possible to perform such a thing? I know that classical 
> ICA results in a separating SQUARE matrix 64x64, the inverse of which 
> is used by topoplot (the columns of A=inv(W) to be more specific).
> In my case however the separating matrix W is rectangular and no 
> matrix inversion can be performed. Can the pseudo-inverse of W be used 
> without error?

Yes, it can be used. No, it cannot be used without error (meaning 
expecting that (EEG.icawinv*EEG.icaweights*EEG.icasphere)*EEG.data 
recovers the original data). The reason is that the number of dimension 
is reduced by PCA before running ICA and there is no way you can recover 
the information lost when performing this transformation.

You can still visualize the component scalp topographies and they are 
meaningful (you simply obtained them applying ICA not on the data but on 
a PCA reduction of the data).

Best,

Arno