I am modeling a task with a mixed block event-related design. It has four block conditions (negative vs. neutral images with predictable vs. unpredictable presentation timings). Blocks are 91 s long, and there are two blocks of each condition type. I modeled these blocks with the BLOCK function, and found some interesting differences within a couple of ROIs.
Now I would like to graph the timecourse of these blocks for each condition, to see if there is anything notable about the response profiles within the ROIs (i.e. is there a big initial activation that tapers off before the block ends, or is it sustained throughout the block). What is the best way to do this? Should I rerun 3ddeconvolve using a tent function (for example, TENT(0,91,15) to get estimates of activation across 15 time windows within the block)? Or should I take the relevant 91 s for each block in the fitts file, and then average the 2 repetitions of the block conditions together? Or perhaps use 3synthesize to extract fit for each condition regressor individually?
I attempted 3dsynthesize on one of my subjects. I added the cbucket option my 3dDeconvolve, with conditions modeled using the BLOCK function. I then ran:
with the 12 in the -select option being the column for the Predictable Negative condition regressor.
Looking through the subbricks of the resulting file, I noticed that while the magnitude of the response was different for each voxel, the shape of the response was the same for all voxels, and seemed to be determined by the shape of the BLOCK function. That is, all the voxels in this time series had activation that ramped up over a a handful of TRs, stayed constant for the rest of the duration of the block, and then tapered back off over a handful of TRs. This made me think that this method won’t give me what I want, which is the ability to look at the shape of the actual response during the block (but with noise and activity from other conditions regressed out). My guess is the using the fitts file has a similar issue - that the shape will be largely determined by the shape of the BLOCK function. This makes me think that rerunning 3ddeconvolve with the TENT function is what I want, since this will allow the data to take any shape, rather than having a particular shape imposed upon it.
Does my thinking here make sense, or am I missing something?
I noticed that while the magnitude of the response was different for each voxel, the shape of the response
was the same for all voxels, and seemed to be determined by the shape of the BLOCK function.
That’s not surprising at all considering you already fixed each condition with the same shape. With a 20s gap, I would simply start with visualizing the fitted time series, and then try TENTzero for each block separately.
TENT would estimate the BOLD response at the onset of each trial. Unless you want to capture something special about the onset, TENTzero is more reasonable and parsimonious.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.