Creating an HRF stim file for rodent GLM

Hello great people of AFNI

I have come to find out that the stim file we have been using for rodent fMRI in 3dDeconvolve isn’t an ideal shape and am needing to create a new stim file. I was told to deconvolve the “real” signal with a double gamma function and was wondering if I could do such a thing with afni. Tutorials I see all seem to create an HRF based on the onset of stimulation, but I believe with how different the HRF is with rodents, those methods won’t work?

What I have is, after 3dDeconvolving with a simple boxcar onset/offset stim file, I drew an ROI in an area that show activation and calculated the percent change. This percent change is assumed to be the “real” HRF… so I have a file called S1.1D. Is it possible to convert this into a cleaned shape that can be used as a stim file?

Furthermore, I wanted to see the effect of different stim files during 3dDeconvolve (eg, gamma and boxcar), so I thought I could set the operator as -num_stimts 2 and input the corresponding stim_files, but when I check each stim_file deconvolution in AFNI, it seems its running GLM again after the first stim file as there’s less “significant” voxels than when I run 3dDeconvolve with the separate stim files individually. Is it possible to have statistics visible in one 3dDeconvolve output with various stim files?

I greatly appreciate any input and your time.

It’s not clear what kind of stimulus timing arrangement you have. Instead of presuming a fixed-shaped HDR, it might be more accurate to estimate its shape using a basis set such as -TENTzero or -CSPLINzero in 3dDeconvolve.

In this paper looking at rat barrel cortex, we used TENT functions and then fit those with a nonlinear model using 3dNLfim. As I recall, the BLOCK function was not a good fit for this data, and gam was better, but none were a terribly good fit. If you look through 3dDeconvolve’s help, you will see about 7 different variations of gamma variate models you can use. You can experiment and see how each looks in the fit time series with -fitts.