I was interested in going after a similar question recently (disentangling
effects of initial stimulus presentation and subseqent feedback). since I was
worried about correlation between the two regressors modeling the two events, I
ran a small simulation that created a design matrix with random intervals
between stimulus and feedback and random intervals between trials (where a
"trial" is a stimulus-feedback pair). What I found was that even with a
distribution of intra-trial intervals that spanned out to 25 seconds (which was
beyond the psychological limits of my study), the regressors modeling the two
responses still showed correlations of at least 0.4 if not more (and this
discounts completely any sort of delay-related activation, which would further
increase the correlation). This leads me to be worried in general about trying
to separately model events in designs with yoked regressors like the one
described by Dr. Michimata.
One caveat is that the distribution from which one samples the intertrial
intervals may make a significant difference - whereas I sampled from a
rectangular distribution, Anders Dale suggests that it is more optimal to sample
from a Poisson distribution. I will post the simulation code if anyone is
interested in trying this. I'd also be interested if anyone else has ideas
about how to surmount this problem.
cheers,
russ
Narender Ramnani wrote:
> Dear Drs Tanaka and Michimata,
>
> If you wish to use basis functions to model activity related to cues,
> instructed delays and probe/response events, I would certainly ensure that
> the delay intervals are as variable as possible. Otherwise, there will be
> some degree of correlation between the cue and the probe events,
> particularly if there are protracted haemodynamic responses that last well
> beyond your 10 sec. In such event, you will not be able to disabmiguate
> between activity related to the cue, to the probe and to the delay interval
> itself. For a description of appropriate methods, you may wish to look at a
> paper by Toni e al. (1999) in Cerebral Cortex. This study has broadly
> similar methodological requirements to yours.
>
> Toni I, Schluter ND, Josephs O, Friston K, Passingham RE. Signal-, set- and
> movement-related activity in the human brain: an event-related fMRI study.
> Cereb Cortex. 1999 Jan-Feb;9(1):35-49.
>
> In combination with methods by Toni et al., you may also wish to consider
> increasing the trial-to-trial variation in the interval between the onset
> of a scan and the onset of the trial. In doing so, you will increase the
> resolution with which you sample haemodynamic responses. I 'think' this is
> summarised in the following reference:
>
> Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R.
> Event-related fMRI: characterizing differential responses. Neuroimage. 1998
> Jan;7(1):30-40.
>
> With best wishes,
>
> Narender Ramnani
>
> At 05:34 PM 2/6/01 +0900, you wrote:
> >Dear SPMers,
> >We have questions about how to apply basis functions to an Event Related
> >fMRI experiment that involves different time course.
> >
> >It is a working memory experiment, and one trial consists of :
> >Stimulus Presentation(6 to 14 sec, depending on different subject group) ->
> >Delay(10 sec) -> Probe Presentation & Response(2 sec) -> baseline(16sec).
> >TR is 2 sec, so that one trial has 17 to 21 scans. We have two subject
> >groups and we are to compare the activation during the delay period between
> >the groups.
> >
> >By searching the SPM mailbase, we come up with the following ideas:
> >1. use $B!H (Bhrf with temporal and dispersion derivatives $B!I (B to
> >model the period
> >of Stimulus Presentation, because this period has different length depending
> >on subject groups.
> >2. Use $B!H (BGamma function with derivatives $B!I (B to model the delay
> >period.
>
> >3. Use $B!H (Bhrf with temporal derivatives $B!I (B to model the Probe
> >Presentation &
> >Response period.
> >Are they correct? Are there any better ways?
> >
> >Second question is about carry-over effects. Our main interest is about the
> >delay period. But if areas active during the delay period are also activated
> >during the Stimulus Presentation period, then this pre-activation would
> >seriously affect the hrf shape of the delay period, so that we would fail to
> >detect activation in these areas. How one could avoid this?
> >
> >Any comments are welcome and we thank you in advance.
> >
> >Satoshi Tanaka
> >Chikashi Michimata, Ph.D
> >Cognitive Psychology Lab.
> >Psychology Department
> >Sophia University
> >7-1, Kioi-cho, Chiyoda-ku
> >Tokyo 102-8554
> >Japan
> >[log in to unmask]
|