[Eeglablist] independent components (artifacts) from data

[arno]

Dear Sameer,

> The information on 
> http://www.sccn.ucsd.edu/~scott/tutorial/icatutorial3.html 
> is very useful for identifying artefacts and finding the projections of 
> each IC on every channel. However, the projections seem to have a +/- sign 
> ambuguity and a straighforward subtraction removes artifacts from some 
> channels but enhances them in others. (Adding instead of subtracting gives 
> the opposite assymetric effect).
>   

Yes, there is a sign ambiguity but it disappears when you multiply the 
component activity time course with the component scalp map. As a reminder

raw_data = sum_over_i( component_scalp_map_i[single_column matrix] * 
component_acitivity_i[single_row_matrix] )

So for a given component, you may invert the sign of both the component 
scalp map and its time course without changing the result of the sum 
above. Removing a component involves setting its activity time-course to 
0 so anyway the component scalp polarity is irrelevant in this case.

> Is there a way to get around this? I have data from 306 channels, and I am 
> not very interested checking channel by channel for each artifact.
>   

With so many channel, you should use the option 'pca', 150 (if you are 
interested in components underling brain activity. FOr simply removing 
artifacts, 'pca', 50 is probably enough. This is command line option for 
the runica command (menu item "Run ICA").

Best,

Arno