Dear List,
I am trying to run a straight-forward GLM on a fast event-related design and getting confusing results when looking at the HRF & temporal derivative (specifically, reasonable results only for the temporal derivative).
I have a simple response task where participants make a response to the onset of movement in a video ("go signal"). This "go signal" occurs between 0.5 and 2 seconds after a static frame appears. My TR is 2 seconds and each trial is 3.5 seconds with a variable time between the onset of the static frame and the "go signal" so the time of each event is jittered with respect to the TR. There are also an equal number of interspersed "Null" trials which serve as a baseline (and help jitter events). When specifying the design, have used the "Multiple Conditions" option and events onsets are indicated at the time of the "go signal" in seconds with a duration of 0.
When I look at a basic Task (all 4 conditions) vs. Rest contrast in individual subject analyses using an F-test:
[1 0 1 0 1 0 1 0; 0 1 0 1 0 1 0 1]
This gives me reasonable task activation (lots of primary visual, motor, subcortical etc.). However this seems to be due primarily to the Temporal Derivative--when I look separately at the T-tests on the HRF and Time Derivative (or look at plots of the contrast estimates for the F-test) I see virtually nothing for the HRF and the expected task activation only in the temporal derivative. This is true for multiple individual subjects I have looked at, and is very robust (i.e. Task activation for the temporal derivative is evident at 0.05 FWE). I understand that significant Temporal Derivative effects in the absence of a canonical HRF effect are relatively uninterpretable and I can't for the life of me figure out what is going on!
A previous post from many years ago regarding auditory cortex activity suggested this is likely due to incorrect specification so the event onsets, however I have checked the onset times repeatedly and also used a different software package with the same condition regressors which resulted in the expected task activation when looking at the canonical HRF. I have tried multiple variations on preprocessing (e.g. with and without slice timing correction) and nothing seems to change this. It seems unlikely to me that this reflects accurately what's going on--subjects respond around 300ms after the "go signal" so there isn't a good reason for the BOLD signal to simply be delayed relative to the task event. Can anyone think of potential explanations or causes for this? Or further diagnostics to help figure out the source? I should mention that I am relatively new to spm, and so even obvious suggestions are welcome.
I'm at a loss and would appreciate any insight!
Thanks,
Katy
|