Hi Torben
Re:
At 07:29 PM 7/4/2001 +0200, you wrote:
>Dear SPMers
>
>Recently I joined, and enjoyed, the SPM-course. At the course Cathy
>Price gave talked about: "The Critical Relationship between the Timing
>of Stimulus Presentation and Data Acquisition in Blocked Designs with
>fMRI" and showed results from a Neuroimage article about the experiment.
>I was quite surprised that this could be an issue at all in typical
>block designs. Reading the article it turned out that ISI=3.2s, whereas
>the figures in the articles used to illustrate the problem used ISI=4s.
>Attached is a ps-file showing the repressors for ISI=3s ISI=3.2s and
>ISI=4s, the figures are a little bit different from the ones in the
>article maybe because i used the HRF from SPM99. The message is however
>clear: at ISI=3.2s the oscillations are hardly visible and at ISI=3s it
>is not a problem at all!
>Price et al. however seem to have observed some effect of sampling
>distributed compared to fixed, but it is hard to se how bias due to
>integer TR/ISI could be an issue in general.
>Torben Lund
I appreciate your position here as there is no doubt that our findings are
surprising.
However, contrary to what you say, I did not base my SPM talk on the data
reported in NeuroImage. In the study reported in NeuroImage, we
demonstrated (to our surprise) that we had more sensitive results, IN TWO
LANGUAGE REGIONS, when we distributed the sampling of the hemodynamic
response throughout peri-stimulus time.
The work I presented on the SPM short course was a follow up study. The TR
was 3 seconds and the ISI was also 3 seconds. By having the same TR and
ISI, sampling occurs at a fixed point in peri-stimulus time. The variable
was the point it was fixed. eg stimulus occurring when data was acquired
from slice 1, 12, 24 and 48. If blocks of stimuli elicit a steady state
response then it shouldn't matter when the data are acquired. If, on the
other hand, there are unpredictable but regular fluctuations in the
response then data acquired at one time point might result in more
sensitive results than data acquired at another time point. This is what
we found, consistent with our earlier observations that distributed
sampling (where data is acquired throughout the peri-stimulus time) was
more sensitive than fixed sampling.
So, based on your simulations, you may not be worried. Based on empirical
data, I am worried and feel compelled to understand what assumptions are
being violated. As I say, the effects are only observed in some brain
regions but they are areas I am interested in for language studies.
Posible explanations are:
1. there is an atypical response in the language areas which results in
higher frequency ossilations in the HR than we generally assume. This
might result if the affected areas are influenced by "top-down" processing.
In other words, responses are being elicited endogenously rather than
purely by the stimulus.
2. There may be movement or noise artifacts that are correlated with TR.
The explanation is not yet clear but I think it is worth investigating in
order to secure the most appropriate design. For the time being, my advice
for studies of language and higher cognitive functions, is to distribute
the sampling even in blocked designs.
Cathy Price
|