[Eeglablist] baseline removal when running 2 ICAs and frequency of interest finding

[Matt Craddock]

On 03/03/2013 11:29, ida miokovic wrote:
> Dear eeglablist,
>
> After reading through many q&a here on eeglab list, I found the
> recommended literature of Groppe et al regarding baseline removal and
> went through it.
>
> My initial plan was to run ICA on the epoched data 1st time, then to
> remove bad trials (not the bad components) and then to run ICA second
> time after which I will hopefully obtain better decomposition. Going
> through q&a, it wasn't clear to me what is the procedure when ICA is
> applied 2 times. This is short description of the steps as I understand
> them at the moment:
> 1. data epoching (without any baseline removal)
> 2. ICA 1st time
> 3. bad trials removal
> 4. ICA 2nd time
> 5. baseline correction on the whole epochs(?)
> And also one more question, if I am interested into finding a certain
> frequency presence in channels and ICs, is Tools --> Component spectra
> and maps good option? In the field of "Freq.to analyze" I would put the
> one of my interest and obtain the components (for example 5 of them) and
> the channel with the highest power of frequency of my interest. I have
> to learn more about dipfit, since I think I will actually need the
> source of the signal of the certain frequency localization...
>
> Thank you very much in advance,
>
> Regards,
>
> Ida

Dear Ida,

The way I read the Groppe paper (this one, I assume: 
http://www.cogsci.ucsd.edu/~dgroppe/PUBLICATIONS/Groppe2009.pdf), doing 
whole epoch baseline correction before the ICA produces the most 
reliable ICA results:

"Eventually it was discovered that IC reliability for these data could 
be improved dramatically by removing the mean of each epoch instead of 
the mean of the 100 ms prestimulus baseline (a common EEG/MEG 
preprocessing step) before applying ICA"

and

"removing the mean of each epoch produced more reliable ICs than 
removing the 100 ms prestimulus baseline for each participant in all 
four experiments listed in the Appendix (48 data sets in total)"

Since trying it out, I've found my own experience with ICA consistent 
with this.

Presumably, after running the ICA and subtracting components etc, you 
could then then change to another, more typical baseline correction 
period (like -100 ms to 0 ms before stimulus onset) where necessary, if 
you're not analyzing the ICA components themselves.

Cheers,
Matt

-- 
Dr. Matt Craddock

Post-doctoral researcher,
Institute of Psychology,
University of Leipzig,
Seeburgstr. 14-20,
04103 Leipzig, Germany
Phone: +49 341 973 95 44