[Eeglablist] Running ICA on multiple datasets

[Tarik S Bel-Bahar]

>
> Hello Michael,
>

​You can do it either way, but if you ICA each block separately then you
would be subtracting different ICs from each block (and that way is
probably also less valid).

It's likely that if ICA across all the periods you will get probably get an
ICA decomposition that more accurately reflects the full range of Neural
and Artifactual ICs in each person's dataset.

The recommended way is to put all experiemental periods that are from same
or similar tasks into one ICA.
Recordings from the same recording session and same participant should be
ICAed together (unless there is a clear and big difference between
different recording periods).

As I mention at the beginning of this note, you could ICA each period, and
if so, you should see similar Neural ICs across each ICA decomposition, if
the task is essentially the same in each block. But my understanding is
that it is unusual to do ICA on sub-blocks from within a recording period.

Also, remember that ICA wants more rather than less data to give you an
accurate decomposition, so that's another reason to not do ICA on the
sub-periods but rather across the periods.

Last, check the recent literature to get a feel if researchers ever do what
you suggested, and under what circumstances.

Remember that ICA for EEG, though well established, is changing and being
updated rapidly, with many variations of ICA for EEG, as well as a range of
other decomposition methods.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20161113/577a6a0b/attachment.html>