[Eeglablist] ICA after PREP pipeline incl channel interpolation

[Makoto Miyakoshi]

Dear Alexandra,

> (Minor question: At this point I do not baseline correct each epoch, is
this recommended?)

Yes, I recommend it.
https://sccn.ucsd.edu/wiki/Makoto's_preprocessing_pipeline#Epoch_data_to_-1_to_2_sec

> Here is where I am losing track. For this example, there are 116
non-interpolated channels and the data set was re-referenced to common
reference before ICA.
So I guess this might be behind the rank reduction, although I am not sure
how runica knows about this (where is such information retrieved?)

It computes rank in pop_runica() line 523 getrank().

> What precisely is used as the input data now? Any first 115 channels or
only the non-interpolated channels?

It's not the channels, but the rank of data that matters. See this.
https://sccn.ucsd.edu/wiki/Makoto's_preprocessing_pipeline#Re-reference_the_data_to_average

> What happens when the resulting (clean) components are projected back
onto channel space if some of the channels are not actually included in the
data set used to calculate the components?

All of your scalp channels will be there, so no worries.
It's just that you have less number of *independent* sources than the
number of channels. Even if you have 1 independent component, it can
project to 100 channels. In your case, you have 99 components projecting to
100 channels (i.e. one less).

Makoto



On Tue, Aug 16, 2016 at 9:18 AM, Alexandra Yvonne Vossen <
a.vossen.1 at research.gla.ac.uk> wrote:

>
> Dear all,
>
> I realise this is a popular question but I am really uncertain what
> happens to my data in my preprocessing pipeline. I am quite new to ICA/PCA
> decomposition.
>
> First, the raw data are run through the PREP pipeline (in EEGLAB 13.5.4b),
> including the (spherical) interpolation of noisy channels.
>
> Then after rejection of very noisy epochs I run ICA (for identification of
> artefact components) with EEG = pop_runica(EEG,'extended',1) %this is
> somebody else's code that I am re-using.
> (Minor question: At this point I do not baseline correct each epoch, is
> this recommended?)
>
> I get the info: "Data rank (115) is smaller than the number of channels
> (128).  Input data size [115,525723] = 115 channels, 525723 frames/nAfter
> PCA dimension reduction,
> finding 115 ICA components using extended ICA."
>
> Here is where I am losing track. For this example, there are 116
> non-interpolated channels and the data set was re-referenced to common
> reference before ICA.
> So I guess this might be behind the rank reduction, although I am not sure
> how runica knows about this (where is such information retrieved?)
>
> What precisely is used as the input data now? Any first 115 channels or
> only the non-interpolated channels?
> What happens when the resulting (clean) components are projected back onto
> channel space if some of the channels are not actually included in the data
> set used to calculate the components?
>
> Many thanks in advance,
>
>
> Alexandra Vossen
> PhD student
> School of Psychology
> College of Science & Engineering
> University of Glasgow
> _______________________________________________
> 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
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20160819/e9f9f6a8/attachment.html>