[Eeglablist] epoching before ICA for continuous data?

Michael Boyle mrboyle at live.unc.edu
Wed Sep 23 09:24:02 PDT 2015

Hey Jianwei,

ICA doesn't care whether or not the provided data is continuous or
segmented or even scrambled along the time dimension, so long as the data
across channels for each time point is left intact. I would recommend
against scrambling your data in the time domain though, as your computed
ICA time-series will be quite difficult to interpret ;). It will depend
somewhat on your preprocessing steps leading up to the point where you are
ready to compute ICA activations. I believe it is commonly accepted that
the best component activations come from reasonably stationary data (a 1Hz
highpass filter usually recommended to achieve this) and reasonably clean
data (reject stretches of data that contain rare and disruptive artifacts
like head movements that contaminate the signal of many channels).

I filter my data as one continuous time-series just to minimize the amount
of artifacts contributed by filtering edges of a time-series. However, for
my cursory data cleaning following filtering, I find it easiest and most
convenient to epoch the data and reject epochs instead of rejecting
continuous stretches of data. That way it is easier to keep track of data
that have been excluded (you only need to retain one epoch number instead
of the index of every rejected sample or the start/stop index of every
rejected stretch of continuous data) and you kind of have a more
appropriately sized standard unit for the amount of data excluded (# epochs
rejected instead of # samples rejected).

With all that being said, that is simply my preference. If you find it
easier/more convenient to work with the continuous data or find something
unsettling in arbitrarily defining epochs (since you have no events) then
by all means stick with the continuous data.

Hope that helps,

On Wed, Sep 23, 2015 at 11:12 AM Amy Cao <amycao723 at gmail.com> wrote:

> Hi All,
> I have one question:
> I am dealing with resting-state EEG data. So I am wondering whether I need
> to epoch my data into small segments before ICA or I could directly use the
> continuous data for ICA.
> Thank you,
> Jianwei
> _______________________________________________
> 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/20150923/8cb50b0b/attachment.html>

More information about the eeglablist mailing list