[Eeglablist] epoched or continuous data analysis

Tarik S Bel-Bahar tarikbelbahar at gmail.com
Wed Mar 23 19:45:34 PDT 2011


Hello!

Good luck and let me know if the below helps. Hoping all is well in Hungary
!


1. You should expect different ic decompositions when you put in different
data.
2. It is better to give ICA more data than less, but this depends on the
bias of the investigator. Ideally you just have many many epochs of the data
of interest,
but that is often not the case.

3. Using the time in between epochs [longer epochs, or the continuous data
for a particular task] may help your task-specific ICA decomposition

4. From my perspective, the ICs you get from ICAing the continuous data are
probably more representative of independent sources in the data. That
being said, you would want the other tasks or time points to be a relatively
similar paradigm, not incredibly different, or considerably more artifactual

5. If you have enough epochs, then your ICA decomposition will probably be
fine.

6. you need at least 30 X (number of channels squared) time points for good
ICA
decompositions. Please check EEGLAB documentation for specific numbers.
Exactly how much or how little data is needed is an empirical question
for future researchers.

7. You probably are only getting one eye blink IC when decomposing your
epoched
data, because your epoched data has only a subset of the total eyeblink
events in your continuous data. It's also possible that the epoched data is
overall
less dirty or less "blinky".

8. PS: Don't forget to visually and automatically clean your data of
bad channels and artifactual time periods, before running the data through
ICA,
whether epoched or continuous. You're better off visually cleaning your
continuous.
You may also need to further clean the data after ICA, and rerun ICA.



On Mar 23, 2011 5:07 PM, "Barkaszi Irén" <barkaszi at cogpsyphy.hu> wrote:

 Dear all,

I ran ICA on my 3-stimulus oddball task with extended infomax algorithm on
continuous (Neuroscan cnt) and epoched (eeg) data too. I had 800 trials
(SOA=1000 ms), epoch length was 1000 ms (-200 ms to 800 ms).  The components
differed in this two analyses (continuous and epoched data, equal data
length). In the epoched data there are meaningful components which are
absent in the cnt analysis. There is one blink component in the epoched
analysis, but in the cnt data several components (more than 10) consist
blink. Does anyone can explain me my result?

Kind regards,
Iren


_______________________________________________
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20110323/13bc30a7/attachment.html>


More information about the eeglablist mailing list