Dear AFNI experts,

I am bit puzzled setting up my GLM as three different basis functions give different results. It would be great help to get an opinion on which of the models are plausible/appropriate for my experiment.

I have a fast-event related working memory task, in which participants remember sounds (i.e., trial onset) for a brief period of time. In the middle of the retention phase, there is a visual cue, which is followed by 5-7 s delay period until participants hear a probe sound. And my main interest is to model the delay period following the visual cue. For this, I set up my GLM to model sound onset (trial onset), visual cue, delay (after the cue), sound probe, and motor response. All regressors were modelled with GAM, except for the delay period after the cue. I tried 3 different GLM's - the delay period with TENT (0-20s), with dmBLOCK, and with GAM, all with stim_times_IM to regress out varying delay period durations.

With TENT basis function, I see deactivation in the auditory region around 0 - 6 s (deactivation peak 2-4 s) and positive activation around 6 - 14 s (peak around 10s), and comes back to baseline. But with GAM and dmBLOCK, I see positive activation in the same region. Of course, I understand the significant effect could arise in different regions with different models, but I expect the direction of the effect should be somewhat similar if I look into a same ROI. For instance, the model with delay GAM function (which models peak around 5s or so) should show deactivation as TENT. But as I see an opposite pattern, I am questioning what would be an appropriate GLM to model the stimulus-free delay period..

Thank you very much in advance!

SJ

P.S. I try to attach a screenshot of TENT and dmBLOCK map, but attachment somehow doesn't work..