[Eeglablist] Data analysis with few channels

Makoto Miyakoshi mmiyakoshi at ucsd.edu
Wed Jun 3 11:36:09 PDT 2015

Dear Emmanuelle,

Here is Makoto's preprocess pipeline

If you re-reference before ICA, don't forget to reduce the data rank by 1.
You can either reject any one channel or use 'pca' option in runica to
specify the rank which EEG.nbchan-1.

Don't forget to clean your data before ICA. Try artifact subspace
reconstruction (ASR) in clean_rawdata(). You can even choose the parameters
so that it keeps the original data length.

Also for ICA make sure that you have sufficiently large number of
datapoints, which channel^2 x 30 or larger when 32ch.

For high-pass, use 1Hz and transition 0.25-0.75Hz. This is very high
compared with other recommendations.


On Mon, Jun 1, 2015 at 1:20 PM, Emmanuelle Renauld <
emmanuelle.renauld.1 at ulaval.ca> wrote:

> Hi all,
> I have been thinking a lot about the best way to analyse my data,
> considering that I have very few channels (8. Thus 7 after re-referencing).
> So far, I do:
> 1. Re-referencing.
> 2. High-pass
> 3. Low-pass
> 4. ICA decomposition
> However: with only 7 channels, I often had very bad result at ICA. Looking
> at the signals, I saw many bugs happening on nearly all electrodes at the
> same time. If I cut the signal around that bug (ex, remove some data from
> EEG.data and EEG.times), the spectrum is usually already cleaner, and the
> ICA decomposition also works better (for instance, alpha frequencies are
> then well separated from the eye blinks and eye movements components).
> So I started thinking about "epoching" my data, but I don't have events to
> do that. Something like cutting my data into windows of maybe 1 second,
> removing windows where the signal is too big, and then computing spectrums
> and ICA. I did it manually, cutting usually 2-3 sections in the data, the
> worst parts, of usually around 10 seconds each. I was supervising the
> results, and it worked, but if I start doing it automatically, I fear that
> the number of windows rejected increase. What do you think would be the
> effect of computing ICA or spectrums on such annexed data? Could it, for
> instance, create false frequencies?
> I have two types of data: at rest, or doing a DDT task.
> Thank you very much!
> Emmanuelle
> _______________________________________________
> 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/20150603/4514ad32/attachment.html>

More information about the eeglablist mailing list