[Eeglablist] ICA of BioSemi data

Arnaud Delorme arno at salk.edu
Tue Oct 9 07:26:43 PDT 2007


Dear Ryan,

yes, Biosemi data must be referenced before running ICA (because Biosemi 
use active electrodes, not referencing imply that you lose about 40dB of 
signal/noise). However the reference does not really matter (Mastoid, 
CZ, average). Pick whichever you feel most comfortable with. We have had 
acceptable ICA decomposition with different reference montages (in 
theory, ICA is not sensitive to the reference since ICA should be 
insensitive to any linear mixture and re-referencing is a linear mixture 
of all channel activities).

The other steps before running ICA on the continuous data involve 1) 
high pass filtering at 0.5 Hz (or higher) and 2) remove obvious 
electrical artifact and paroxysmal activity (as when the subject move or 
scratch his head).

Applying ICA to the segmented data is acceptable too. Please refer to 
this message:

http://www.sccn.ucsd.edu/pipermail/eeglablist/2005/000885.html

Best,

Arno

Ryan Miller wrote:
> hi all,
> I am attempting to do an independent component analysis of 64-channel 
> BioSemi data.  I am wondering if the data should be referenced (e.g. 
> to the average) before the ICA is run.  Also, what other steps should 
> be completed before doing an ICA (e.g. 
> filter?  baseline-correction?).  I am planning on doing the ICA with 
> segmented data.  Thanks for your suggestions,
>
> Ryan
>   



More information about the eeglablist mailing list