[Eeglablist] nonstationarity issue

[Makoto Miyakoshi]

Dear Kathleen,

I have never used BCILAB so may not be helpful, but since you haven't
received any comment so it may be better than nothing...

Preprocess the data with ICA so that extract (source-resolved, Scott says)
signal from scalp mixed EEG. Basically, as long as your experimental design
and ICA are successful, you should be able to extract decent signal there.
Non-stationarity of these source-resolved EEG signals should be apparent if
you perform event-related or block-separated analysis, suggesting that your
experimental control over some cognitive conditions (i.e. independent
variables) was successful, which should be separated by potential
non-stationarity of artifacts because they were decomposed and removed by
ICA. This logic should always work, no?

Makoto



On Thu, Feb 25, 2016 at 8:07 AM, Kathleen Van Benthem <
kathy_vanbenthem at carleton.ca> wrote:

> Hello List,
> I have looked for a BCILAB forum but couldn't find one, so I have posted
> here.
>
> We have recently had a paper rejected where we classified low and high
> mental workload states (a match to sample :easy vs hard task) using
> *BCILAB*, the CSP (between 7 and 30 hz) and LDA .  We analyzed epochs
> based on the onset of standard and deviant tones- also labelled as per the
> low and high workload task conditions (participants wore earbuds and were
> told to ignore the tones while completing the task).
>
> We used a 128 channel system and decided to use all the channels in the
> classification analysis.  We achieved a sig. better than chance
> classification rate for all four different classification schemes (based on
> what tone type and epoch segment we used).  We counterbalanced between
> subjects regarding whether they started with the easy or hard condition of
> the task.
>
> The rejection was based on nonstationarity issues.  Apparently the
> dynamics of the brain are so wildly fluctuating that you cannot draw
> conclusions any time you introduce a time series.  My thought is that I
> should demonstrate that the mean and variance of some other features of the
> EEG signal were *not* significantly different between conditions- eg. eye
> artifacts.  Thus when I show differences between my workload conditions for
> other EEG features I can make a case that it was the task manipulation that
> likely resulted in the differences in EEG signals, and our classification
> rates were not simply a result of nonlinear and nonstationary EEG.
>
> Questions;
> Has anyone else run across this problem with nonstationarity in repeated
> measures?  and how did you solve it?   Are there methods within *BCILAB*
> to deal with time series issues?
>
> ps. we could not alternate between easy hard easy hard easy hard etc...
> within subjects or you would lose the stable mental workload state you were
> trying to induce.  For example, in real life if you were engaged in a task
> that went from easy to hard repeatedly in a short time frame, IMHO, you
> would essentially be in a high workload state the entire time.
>
> Any and all suggestions are welcome!
>
>
> ​
> Kathleen Van Benthem Ph.D., AGE-WELL HQP
> ​Instructor FYSM 1607C​
> Postdoctoral Fellow, ACE​ Lab
> VSIM Building, Carleton University
> kathy.vanbenthem at carleton.ca
> Phone:  (613) 355-2515
>
> _______________________________________________
> 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/20160404/b3d1c0f4/attachment.html>