[Eeglablist] Time-frequency analysis (subtraction first or analysis first)

[Stanley Klein]

Dear EEGlist, First I'd like to comment on how it is cute that this
discussion has shifted from the original point of the subject heading, but
the same subject heading is relevant to the issue of whether to first
subtract off the ERP before doing the time-frequency analysis.
Stan

Dear Brian,
On the gamma/microsaccade connection. There was a poster at the Cognitive
Neuroscience meeting a few days ago by Shlomit Yuval-Greenberg titled
Reassessment of the origins of Induced Gamma Band Responses – A single trial
analysis. I didn't see the poster but I heard that that their results do not
apply to ECoG or  LFP, only to EEG. It will be interesting to see discussion
of this interesting topic when their paper gets published.

On the topic of 'evoked' vs 'induced' responses. I'd like to suggest that
when one has a really interesting result, say about gamma, it would be
useful to the reader to show the data in more than one way. The following
are some possibilities:

1) Show the old fashioned ERP(t) averaged across trials and subjects.
2) Show the TF plot of the ERP.
3) Show the TF plot of the combined 'evoked' and 'induced' as most folks on
this list suggest
4) Show the TF plot after subtracting off the ERP. This has the advantage of
possibly emphasizing feedback pathways where the cognitive based time delays
can be more variable than the bottom up stream. (Does this make any sense?).


One reason to show ERP(t) is that the filter that is used typically has a
very broad bandwidth so one sees something close to the true time course of
the response. When one sees (as one often does) some very sharp features in
the temporal response one gets excited. The problem with the TF plot is that
one typically uses a wavelet filter with more than one cycle so one distorts
(adds wiggles to) sharp temporal responses (like a microsaccade?).

The big advantage of methods 3 and 4 are that they capture the
non-stationary aspects of the response. It makes a lot more sense to average
TF plots across subjects than to average ERPs across subjects because
individual differences in the cortical folding can easily produce
differences in the scalp potentials. However the folding makes any type of
scalp based averaging (including TF plots) across subjects less than ideal.
That's why people try to get to source space.
Stan






On Wed, Apr 16, 2008 at 2:22 PM, Brian Roach <brian.roach at yale.edu> wrote:

>  There is an article by Truccolo et al in Clinical Neurophysiology
> (Trial-to-trial variability of cortical evoked responses: implications for
> the analysis of functional connectivity) which examines such a subtraction
> of the evoked response.  I believe the idea is that your averaged ERP does
> not exactly represent all potential evoked activity in a given trial.
> Therefore some residual evoked activity will survive a "single trial minus
> ERP" routine.  ERSP data contains both evoked and induced power, which are
> tough to separate without making some assumptions about the data.  I don't
> think that you can subtract the ERP from every single trial, subject it to
> TF analysis, and say that the resulting ERSP map shows phase varying (or
> induced) responses *only*.  To do that, you would have to assume that the
> ERP subtraction accounted for exactly all of the evoked power activity in
> every single trial.
>
> I was interested in your last comment Stan - do you have any references
> about saccades and gamma?
>
> thanks,
> Brian
>
>
> Stanley Klein wrote:
>
> It looks like there is some consensus on whether to subtract first and
> then the TF or vice versa. That's nice. [On the other hand subtracting first
> is a nice way to get rid of ERP, but there are better ways, as described
> next.]
>
> Andrei, I'm not sure I understood your last comment or question, but I
> have a related question. Whenever one does time-frequency power plots I
> would think that one should ALWAYS first get the time locked average and
> subtract it off  of all the individual trials. Then one could do a TF plot
> of each. How many on this list do that?  I suspect that people mix together
> the standard evoked response and also the phase varying response. Why do
> that since its so easy to show the the two TF plots separately.
>
> Also I've heard rumors that saccades and microsaccades are responsible for
> most EEG gamma oscillations. So one should also put those events into a
> separate category too. Too bad things are complicated. But it makes life
> interesting.
> Stan
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20080416/510de8f0/attachment.html>