[FieldTrip] Reject components on unfiltered data

[Matt Craddock]

On 28/10/2013 15:40, Raquel Bibi wrote:
> Hi Eelke,
> I often wonder if it is better to filter or not filter prior to ICA.
> While Carmen never asked, I wonder what is the 'Fieldtrip' Donders' view?
>
> Best,
>
> Raquel
>

Hi Raquel, all,

I can't speak for the Donders' view, but a high-pass filter (0.5 Hz, 
typically) before ICA often improves the quality of decompositions, 
since it improves the stationarity of the data by removing slow drifts. 
It's fine to then apply the weights from the resulting decomposition to 
the unfiltered dataset. I'd be cautious about applying too drastic a 
lowpass filter, as it tends to make the data from the various channels 
more similar and thus less independent from each other - it sometimes 
reduces the rank of the data, and ICA can behave a little weirdly with 
data which isn't full rank. Another handy thing to do is remove the mean 
of the whole epoch before running the ICA - see this paper from Groppe 
et al. (2009) http://dx.doi.org/10.1016/j.neuroimage.2008.12.038

Cheers,
Matt

-- 
Dr. Matt Craddock

Post-doctoral researcher,
Institute of Psychology,
University of Leipzig,
Neumarkt 9-19,
04109 Leipzig, Germany
Phone: +49 341 973 95 44