[Eeglablist] ICA on baseline corrected epochs

[Arnaud Delorme]

Dear Jordi,

according to our test and a paper by David Groppe, if the baseline is too short (~100 ms) ICA component reliability degrades critically. This is because you are introducing non-linearity by removing the baseline (if there are common generators that could be isolated by ICA and project linearly to each channel, you are making it harder for ICA by subtracting a constant from each channel activity - since this common source cannot project linearly to all channel any more).

Groppe, D.M., Makeig, S., & Kutas, M. (2009) Identifying reliable independent components via split-half comparisons. NeuroImage, 45 pp.1199-1211.

I would personally recommend a baseline period of 1 second.
Hope this helps,

Arno

On Oct 30, 2010, at 9:54 AM, Jordi Costa Faidella wrote:

> Dear EEGlab users,
> 
> I have a question regarding the use of ICA on baseline corrected epochs. Playing around with a dataset, I realized that the ICA doesn't retrieve exactly the same results when I apply it on a continuous EEG file or in concatenated, baseline corrected epochs. At the beginning I thought it was only a matter of using different number of data points, or the information in the datapoints per se. Thus, I tried the following: I extracted and concatenated epochs from a single dataset in two different ways, one without and one with baseline correction. The reason I did that it was because, as far as I now, when data is baseline corrected we only can look at the differences in scalp topography between the baseline period and a given time period, but not the "real" topography that was recorded in the original continuous file. The results I obtained differed slightly in components with high weights, like artifacts etc., but differed quite a lot in smaller ICs. Thus, I'm wondering whether!
>  it would be more correct to apply ICA on a continuous dataset, or at least in concatenated epochs without baseline correction, than to apply it on a set of single baseline corrected epochs.
> Does anyone know what is better? Or is it the same and my results were due to an error (unknown to me)?
> 
> thank you all,
> 
> Jordi
> _______________________________________________
> 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