Aaron Trachtenberg wrote:
Hi, I am trying to design an event-related memory encoding task. Simply put, the subject either sees familiar or novel pictures (the familiar ones having been shown outside of the scanner before hand). After scanning, recognition for the pictures is tested. I am interested in the novel > familiar contrast but also subsequent memory (i.e. which images were remembered). I have two questions:
- For planning the schedule, I will use optseq2 with an input of 2
evs (novel and familiar). It is my understanding that these will be optimized for maximum efficiency. But if I post-hoc separate the novel events based on if they were subsequently remembered or not, could this not result in a very inefficient design? In other words, is an optimized schedule output from optseq2 remain fairly optimal under a variety of possible post-hoc sortings? Is there anything special I can do, knowing that I plan to use post-hoc sorting?
There's nothing you can do in optseq. I could imagine some elaborate simulations you could do, but I doubt it's worth it. I imagine that, when all is said and done, it won't make that much difference.
- Having worked through the examples on the internet, is my
understanding that a stimulus event always occurs at the start of a TR correct? How does this affect sample bias? Is there away to have a stimulus onset at different points in the TR? I may be confused about this point, but it is my understanding that you don't want the stimulus occurring at the same time with respect to every TR or else every slice will always be at a consistently different position in the HRF?
This is a slightly complicated questions with several facets. First, you can instruct optseq2 to present stimuli off-TR by specifying a sub-TR value for dPSD. This has to be an integer divisor of the TR (eg, 1 or 0.5 for TR=2). This allows optseq to present off TR (some presentations will be, some won't). When you do this, you'll notice that your variance reduction factor will drop by an amount about equal to the divisor, meaning that this design is less efficient. This is because optseq assumes an FIR design, and a sub-TR dPSD requires the FIR time points be sub-TR (so you have many more parameters your trying to estimate, and so less power/efficiency). However, if you plan to assume a shape to the HRF (eg, gamma or SPM HRF), then it will not actually hurt your power to go sub-TR.
doug
Any help on these questions, or general insight into this kind of experiment would be great. Thanks a lot Aaron Trachtenberg DPhil candidate, FMRIB, University of Oxford _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer