[Eeglablist] Re: ICA-based removal of eye artifacts

[Scott Makeig]

If you spatially filtered the twitch artifact using the ICA weights not 
trained on the twitch, then the weights would 'share' (in some fashion) 
the artifact, and the back-projection of all or some of the resulting 
component activations would indeed spatially smear out the twitch into 
other channels. If the resulting data was used for averaging of some 
osrt, then the 'noise' effect on the average might (or might not) be 
acceptable. Best would be to focus on periods of relatively clean data 
('GIGO').

Scott

Mante Nieuwland wrote:
 > Thank you very much for answering my question (so swiftly and
 > extensively!). The dilemma that i face(d) turned up because the Jung
 > et al (2000) papers in Psychophysiology and Clinical Neurophysiology
 > introduced different procedures. I had doubts about using all the
 > data, because data recorded during 90 minutes of listening to stories
 > would violate the stationarity assumption even more than the data
 > from a selective attention task. But I understand from your answer
 > that using all data for artifact correction is acceptable, as long as
 > the data is optimalized for ICA ('pruned' from paroxysmal artifacts).
 >
 >>>> (Note: the resulting weights might be applied to the removed
 >>>> portions of the data as well, with the strong caution that
 >>>> the  resulting activation time courses will be noisy
 >>>> and less representative of the actual component process
 >>>> activity).

 > I think I have already encountered this problem. If for instance a
 > strong muscle twitch is present at one sensor(lets say EOGV1), during
 > the pruning it will be removed before training the ICA, but during the
 > correction-procedure this muscle twitch can turn up (though less strong)
 > at different sensors than the original sensor. Afterwards, one might
 > reject an epoch containing this artifact, eventhough the original epoch
 > would never be rejected since the artifact was present at only one
 > sensor (and one could have rejected a one channel-epoch only). So these
 > muscle twitches (and all other paroxysmal artifacts) can become
 > 'smeared' over sensors. Is this what you meant by the note?
 >
 > Regards, Mante> 
> Drs. Mante S. Nieuwland
> University of Amsterdam
> Dept. of Psychology (PN)
> Roetersstraat 15, kamer 614                     
> 1018 WB Amsterdam                                    
> The Netherlands 
> mnieuwland at fmg.uva.nl
> phone:   +31-20-5256808
> fax:   +31-20-6391656