[Eeglablist] ica-based artifact removal

[Mante Nieuwland]

Hello,
Another question about ICA-based artifact removal. In my dataset blinks and horizontal eye movements were captured very nicely into independent components. But it is likely that this dataset also contains some vertical eye movements (just as any other dataset). Blink artifacts result from conductance difference due to movement of the eyelids, but artifacts from vertical eye movements result from the rotating eye-dipole. But any vertical eye-movement is accompanied by a movement from the eye-lid, looking up without adjusting the eyelid wouldn't get us anywhere in terms of sight gain. This would mean that these components are not completely independent, resulting in a less neat set of components. Is there anybody that has encountered similar problems?

And one more question: I have tried ICA on a 1-hour dataset full of eye movements, namely a sort of eye-tracking calibration task requiring a subject to follow a crosshair changing location every 1.5 sec within a 3x3 maze. This ICA did not return a clear blink-component or horizontal/vertical eye movement-component. Possibly the interrelation between looking up or down and moving the eyelid has led to these problems. But there is another issue: when I tried to replicate the Jung et al (PsyPhys, 2000) findings on 10-s parts of my own data, the resulting blink-component was most clear when this 10s secs contained only 1 blink, but when these 10 seconds contained significantly more blinks there were several components with blink-activity. This suggests that the amount of variance in a given data-set accounted for by one kind of activity can lead to unwanted results, just as an ICA of the eye-tracking task data did not give satisfactory results, while an ICA on a dataset of the same length from a language comprehension task full of blinks did give good results. Not enough or too much variance would then lead to erroneous components? Furthermore, in the Jung et al paper (Clin.NeuroPhys, 2000) the subjects were reported to make a lot of small horizontal eye movements (although this was not part of the experimental instruction, as would be in a tracking task, so whether movements are voluntary could be of influence as well), and these movements fitted nicely into 1 component. I am not sure how to explain these discrepancies, so I would be very thankful for any helpful comments,

Many thanks in advance,

Mante Nieuwland

Drs. Mante S. Nieuwland
University of Amsterdam
Dept. of Psychology (PN)
Roetersstraat 15, kamer 614                     
1018 WB Amsterdam                                    
The Netherlands 
m.s.nieuwland at uva.nl
phone:   +31-20-5256808
fax:   +31-20-6391656
see also: http://users.fmg.uva.nl/mnieuwland