John Herrington wrote:
I'm new to optseq and to rapid E-R designs generally. I have a two questions about it. If it helps, here's my (tentative) design - I'm creating a task with 5 conditions (including rest) with a TR of 2320 ms and trial SOA of 4640 ms (my stimuli are on screen for 1 sec of this time).
- I've been trying to get clear on the logic behind optimizing rapid
E-R designs. My read of this literature has led me to the conclusion that there are two overlapping ways to do it: A) vary SOA from one trial to the next, without necessarily paying much attention to trial ordering, and B) vary the trial ordering in a maximally efficient way while keeping the SOA constant from one trial to the next (this is in effect a "jittering", if one considers the variations in time between trial types that results from the pseudorandomization). One can of course combine A and B, but if my logic holds and read of the literature is correct, you could stick with one or the other and still wind up with a sufficiently optimized design. Is this logic and conclusion correct? If so, your program creates a time series via approach B and not A, yes?
optseq forces you to combine these strategies. You have to have some null time because it uses an FIR model. In general, you give it as much total null time as you would for any other stimulus (though it will not be constrained to create null events of any particular duration).
2)I'm somewhat familiar with the concept of efficiency in the context of rapid E-R designs, but I haven't found a good resource that tells me what to do with that statistic, as implemented in your program. In other words, what is a passable efficiency threshold for optseq2? I imagine that, like many parameters one selects in MRI, it might depend on any number of factors... but is there any heuristic to apply to the efficiency number generated by your program to figure out what's good and what's bad?
Your question is more about power than efficiency. Obviously, if you don't have an effect, no amount of efficiency will be enough. Having said that, I usually look at the variance reduction factor (VRF) rather than the efficiency (though they are related). The VRF shows up in the t test as:
t = avg/sqrt(var/VRF)
I usually shoot for a VRF of 15-30 to see good activation in primary sensory or motor areas.
FSL has a power analysis tool (though they refer to it as efficiency:). You can generate schedules with optseq and then feed them into FSL to do the power analysis.
doug
Thanks,
John
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer