[Eeglablist] ICA and PCA

[Davide Baldo]

Dear all,

I went through many posts on the EEGLABLIST about ICA and weather or not to
perform PCA before ICA. But I still have some doubts. Some time I read
posts in which PCA is suggested before ICA and some times is suggested to
avoid it, so I am still a bit confused.

Let me first describe how do I record EEG and which are the first data
processing steps:

> I collect EEG data with a Biosemi system (64 EEG channels and 4 EOGs,
srate: 1024Hz).

> All the 68 channels are referenced to average reference and then filtered
to remove noise (HP 1 Hz LP 45 Hz, I am not interested in ERPs that s why I
use an High Pass filter of 1 Hz)
Note: The average reference is computed as the average of the 64 EEG
channels only, I do not include in the average reference the 4 EOGs
channels.

> Data are divided into epochs

> At this point I would like to run ICA in order to remove (mainly)
eye-movements and eye-blinks

My questions are:

1) shall I reduce the data dimension from 68 to 67 performing PCA before
ICA (because the average reference reduced the rank of the data)?

2) shall I reduce the data dimension more than that? For example I could
check how many components (PCA) I need to explain the 98% of the variance
of my data

3) if I have very noisy channels shall I just remove them before computing
ICA and PCA?

4) would you include the 4 EOG channels when running ICA? I read that is
fine as far as each channel has the same reference so that should be okay.

Thanks a lot for your kind help,

Davide
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20130911/7bd6a382/attachment.html>