[Eeglablist] removal of event-related (time/phase-locked) variance from ongoing oscillatory activity (cf. Makeig '93)

[Alexander J. Shackman]

Dear Jeff --

Thanks for the very helpful follow-up to Adrian's response. In my
original post, I proposed a procedure similar to that you have
detailed based on K&P95, in which the bandpassed ERP was computed and
then regressed from the individual EPOCHS (separately for each
combination of subject and condition) in order to remove the
event-locked variance, and allow us to compute a more pure measure of
condition-related changes in oscillatory activity. Following
regression, an FFT could be computed and the average power (uV2/Hz)
computed for the band of interest -- not unlike the squaring you
described. It seems to me that the regression procedure would be more
robust to violations of the ERP homogeneity assumption (i.e., that the
time-locked signal has stationary amplitude and latency) than the
subtraction procedure. Further, the residuals would be orthogonal
(uncorrelated) with event-locked (time-/phase-locked) activity.

My major concern is whether the assumptions underlying this regression
procedure (or simple subtraction) are substantially violated by work
showing interactions between event-locked and on-going oscillatory
activity, such as phase re-setting or related phenomena. A related
concern is whether the event-locked signal is likely to be stationary
enough to allow this procedure to grab most of the event-locked
variance?

Any advice on these points would be gratefully appreciated!
Alex




On 12/18/05, Jeff Hamm <j.hamm at auckland.ac.nz> wrote:
>    If I understand you correctly, the following two papers may be  of
> interest. If you want to separate out the phase locked activity,  and just
> look at the non- phase locked activity, then Kalcher and  Pfurtscheller
> (1995) suggest the following procedure.
>
>
> Step 1: filter the individual EPOCHs to the band of interest.   Step 2:
> average the EPOCHs into a "filtered ERP".
> Step 3: subtract this average ERP from each individual EPOCH.
> Step 4: square these difference values
> Step 5: average the resulting squared deviations.
>
>
> Salmelin and Hari (1994) suggest an almost identical procedure,  with the
> only change being that in step 4 you simply take the  absolute volage rather
> than square the voltage (they call this  temporal spectral evolution
> analysis).
>
>
> Both of these procedures assume that the ERP is unchanging  from trial to
> trial in both amplitude and time.  If the ERP  response varys, then the
> subtraction of the grand average ERP  will leave a residual component
> behind.  Although the squaring  procedure of K & P would appear to magnify
> this residual, the  non-phase locked activity (which should be unaffected by
> the  subtraction) will be greatly increased.  As such, the resulting
> "waveform" will reflect the non-phase locked activity more than  any
> residual ERP to a greater extent than the absolute voltage  method of S & H.
>
>
> Perhapse using a de-correlation procedure in step 3 (de-correlate  the
> EPOCHs and the ERP) would be more effective to deal with  any trial by trial
> variation in the individual ERP response.  Of  course, this would continue
> to assume that there is no temporal  variation in the ERP response (as do
> the other procedures), but  should remove the assumption of non-variation in
> amplitude.   I'm not aware of anyone using decorrelation rather than
> subtraction, and there may be some concerns about this as a  procedure (I'm
> just thinking of it from reading your message).   Anyway, hope these are of
> some help.
>
>
> - Jeff Hamm
>
>
> Kalcher, J. & Pfurtscheller, G. (1995). Discrimination between  phase-locked
> and non-phase-locked event-related EEG activity.  Electroencephalography &
> Clinical Neurophysiology, 94, 381- 384.
>
>
> Salmelin, R. & Hari, R. (1994). Spatiotemporal characteristics of
> sensiomotor MEG rhythms related to thumb movement.  Neuroscience, 60,
> 537-550.


--
------------------------------------------------------------------
Alexander J. Shackman
Laboratory for Affective Neuroscience | Waisman Laboratory for Brain
Imaging & Behavior
University of Wisconsin-Madison
1202 West Johnson Street
Madison, Wisconsin 53706

PH: +1 (608) 358-5025 (cell)
FAX: +1 (608) 265-2875
EMAIL: ajshackman at gmail.com
WWW: http://psyphz.psych.wisc.edu/~shackman |
http://brainimaging.waisman.wisc.edu/~shackman/