[Eeglablist] Multiple sessions - same condition - per subject

[Makoto Miyakoshi]

Dear Chadwick,

So you do BCI at Keio. I'm not a BCI person, so I could be ignorant of
common practices in the field. Let me give you some general ideas.

> How do I perform a between-subjects statistical analysis on
independent components from data that are obtained over multiple sessions
within a subject?

If that multiple sessions involves multiple electrode cap/net setup then I
don't recommend running ICA on the concatenated data since their channel
locations across sessions would not be identical. If it was done on the
same day with the same cap setup it's fine.

> I can treat each subject*session as an independent subject so I have
62 subjects (13ss * 5 sess - 3 noisy sessions) instead of 13. I think this
is wrong because it gives me too much statistical power and some
subjects are underrepresented.

You should not double count subject in that way.

> I could then concatenate activations across sessions into a
super-session. Could I then use the average weights and dipole centroid for
the super-component?

That does not seem straightforward when you can concatenate your raw data.
Try concatenation first if it is allowed.

Makoto

2013/3/4 Chadwick Boulay <boulay at bme.bio.keio.ac.jp>

> Dear all,
>
> I'm sorry to send two messages on the same day. I have a question
> related to the IC clustering discussion.
>
> How do I perform a between-subjects statistical analysis on independent
> components from data that are obtained over multiple sessions within a
> subject? This was a BCI study and each session was a near-identical
> replication of the previous (i.e. same 'condition' in each session).
> This was a P300 study so I do not expect there to be much of a learning
> effect and I do not expect the P300 'source' to change across sessions.
>
> I tried pretending all data came from the same session then running ICA
> on the concatenated data and I ended up with 5 ICs for eye blinks (5
> sessions) with each of those ICs representing mostly a subset of trials
> from a single session.
>
> I can treat each subject*session as an independent subject so I have 62
> subjects (13ss * 5 sess - 3 noisy sessions) instead of 13. I think this
> is wrong because it gives me too much statistical power and some
> subjects are underrepresented.
>
> I can choose a set of ICs from one session then recalculate activations
> and concatenate trials, but which session do I use?
>
> Finally, I can cluster ICs within-subject in a manner similar to the
> 2012 EMBC IEEE Conf Proc by Grandchamp et al
> (http://www.ncbi.nlm.nih.gov/pubmed/23367475). Then, I could manually
> reorder the ica weights so that each session's ICs are ordered according
> to their cluster-membership then recalculate activations. I could then
> concatenate activations across sessions into a super-session. Could I
> then use the average weights and dipole centroid for the super-component?
> Can someone please share with me the modified clustering code to
> constrain the clusters to only include one component per session?
>
> Thank you,
>
> Chadwick Boulay
> JSPS Postdoctoral Fellow
> Keio University
> Yokohama, Japan
> _______________________________________________
> 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
JSPS Postdoctral Fellow for Research Abroad
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/20130307/480ce2ec/attachment.html>