[Eeglablist] Interpolation and ICA

Andreas Widmann widmann at uni-leipzig.de
Wed Jan 6 04:38:59 PST 2010


Dear Bernhard and Arno,

imho, it is also relevant (a) why you want to interpolate/the type of artifact and (b) whether you want to interpolate the channel for the whole dataset or only parts/trials/epochs.

(a) E.g. for unplugged or broken active electrodes containing noise and no signal I would prefer removing the channel and inserting an interpolated channel after ICA. Arno, would it be safe to feed (partially) noise only channels into ICA?

(b) If you interpolate ONLY PART of the dataset/epochs/trials AFTER ICA computation, imho ICA solution is no longer strictly valid for the dataset. Did I misunderstand something here?

Two comments:
Using the sphspline plugin instead of eeg_interp you should be able to interpolate the channel for ICA components too.

Spherical spline interpolation does not necessarily introduce non-linearity (the name is misleading). E.g., the back-projected dataset with a channel interpolated for all components is identical to the dataset with the channel interpolated.

Best,
Andreas

Am 05.01.2010 um 01:20 schrieb Bernhard Meyer:

> As I need to use the algorithm eeg_interp(EEG, Channel, 'spherical') for
> channel-interpolation in combination with ICA I found a possible
> contradiction regarding the question what to apply first, ICA or
> interpolation.
> 
> Arno Delorme wrote in 2006 (Archive): 'Note that it is not recommended
> to perform interpolation before running ICA. The attached function will
> take care of you ICA information when you perform interpolation.'
> 
> Andreas Widmann wrote in 2009 (Archive): 'However, be aware that a
> possibly existing ICA weight matrix will be invalid as only part of the
> dataset is interpolated!'
> 
> In my opinion I would run ICA after interpolation. On the one hand I
> am afraid to have invalid ICA-data in the end, especially if I
> interpolate just a part of the epochs of the channel. On the other hand
> I want to have all channels as clean as possible, otherwise I could lose
> degrees of freedom for useless data.
> I hope the EEGLAB-community can clarify that step.
> 
> Best,
> Bernhard Meyer
> 
> _______________________________________________
> 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