Need help determining proper setup/settings in FEAT to analyze fMRI data from an auditory pitch discrimination experiment that was conducted with a sparse sampling experimental design. I’ve been running FEAT with default settings and most of the resulting post-stats show no activation at all (although default settings have produced a result on a very limited number of participant’s data). In runs of FEAT where I get no activation, I know this to be in error because I do show activation running MELODIC on the same data, and I have rich behavioral data as well showing that something was in fact going on in these sessions.
Details of the experiment:
Experiment consisted of a pitch discrimination task in which subjects were presented with two pitch stimuli (tones), one after the other. The second tone was either the “same” or “different” than the first tone (the second tone stimulus varied up or down in pitch relative to the first, but the subject had only to respond that the tone was different in general from the first.) Subjects then recorded their response of “same” or “different” via a button press.
- The experiment consisted of 58 TRs of 9s duration each.
- E-Prime was used to present the auditory stimuli, and was triggered automatically by the TR pulse from the MR machine.
- Each TR breaks down chronologically like this:
o At the start of the first TR, t=0, the MR pulse triggered E-Prime to start with an inactive phase of variable time (see 4th bullet).
o The MR then acquired one volume over a 5s period while E-Prime was inactive. Note here that this first volume is devoid of any experimentally relevant activation because there was no preceding auditory stimulus – how should this be accounted for in FEAT? Should I actually delete this volume or just adjust my timing file to compensate?
o After 5s of MR scanning, the MR machine became inactive/silent until the beginning of the next TR (so an active period of 5s and an inactive period of 4s in each TR).
o E-Prime became active from its “sleep” after a variable time period, which was at either 5, 5.5, or 6s post-MR pulse induced “sleep.” E-Prime and the MR were never active simultaneously. These variable inactive periods, although random, are known and have been properly accounted for in the 3-column format timing file.
o E-Prime becoming active was the Event Onset in each TR, because it is at that time that the first tone was played. The first tone stimulus lasted for 0.5s, followed by a silent period of 0.5s, followed by the second tone stimulus lasting 0.5s, and finally another 0.5s silent period. Based on this timing, my 3-column format timing file has the Event Onset being the presentation of the first tone, with a 1.5s total duration. Note that my Event Onset times actually have been corrected according to the sparse sampling FAQ instructions on the FSL website (http://www.fmrib.ox.ac.uk/fslfaq/#feat_sparse). So, I adjusted the timings by adding 2s to each Event Onset (9/2-5/2=2s). My experimental design is quite similar to this example in general.
o After the presentation of the tone stimuli, there was a 1s response window for the button press.
o The E-Prime aspect of each TR, which included the presentation of the auditory stimuli along with (2) periods of silence and (1) response window, lasted for exactly 3s. So, taking into account the variable “wait times” following E-Prime sleep, the E-Prime aspect could’ve occurred from 5-8s, 5.5-8.5s, or 6-9s during each TR. If the E-Prime sequence finished before the end of the 9s TR (like at 8 or 8.5s), it would just become inactive and await the MR pulse at the beginning of the next TR to start the whole sequence over again.
As I said, I’ve adjusted the Event Onset times from each TR according to the FSL FAQ post on analyzing sparse sampling datasets (link above). Other than that, I’ve tinkered with a few of the settings, like reducing the “High pass filter cutoff” from 100s (default) to 50s, and changing the convolution settings; I’m thinking it is probably in the convolution settings that most of the adjustment for my experimental design must take place. So far, I’ve kept the Convolution setting on Gamma and the Phase at 0 (both default), but changed the Stddev from 3 to 1.5s and the Mean lag from 6 to 4s. All of these changes have provided mixed results thus far.
My question, then, again: am I on the right track with the settings adjustments, or is there some other area that I should consider changing to best analyze my fMRI data?
Thank you, and please let me know of any questions/clarifications that I can address.
Charlie
|