Thanks, a lot, Colin. I think I will give it a try with FIR, but I also found some more information about FLOBS: apparently it can be used to get individualized HRFs.
I wonder, though, for both FIR and FLOBS, whether there is a problem with performing higher-level statistics on measures derived from this "individualized" HRF? E.g., if I derive a measure of "width" by using the full-width at half-max of the HRF response, can I plug it into a second-level analysis? (I am investigating specific ROIs.)
Below is an email re: FLOBS for individual HRFs from a few years ago.
-----
Hi Stefano,
You can use FLOBS to estimate HRFs for individual subjects. You will need a task that produces a reliable and predictable primary sensory activation. A flashing checkerboard is probably best, though, depending on the task, it may be possible to extract the HRF from the task itself, as long as you are careful to avoid circularity. For optimal estimation make sure your stimulus/checkerboard varies in duration but not in intensity. Enter the stimulus regressor into FEAT with FLOBS as the convolution. Turn on F-test and run FEAT.
Create a mask of the primary sensory region (i.e. visual cortex) and multiply the mask by the fzstat1.nii.gz, then threshold and binarize. Take those voxels, multiply by the parameter estimate map (pe1.nii.gz) for flob1, and average the parameter estimates for those voxels. Repeat for pe2 and pe3. Then do a weighted sum i.e. HRF = pe1*flob1 + pe2*flob2 + pe3*flob3. This will give you a customized impulse response for each subject. For more details see Grinband et al, Neuroimage 2008.
cheers,
jack
|