This falls into the "I love AFNI but want to do SPM-style analysis" category...

I have been playing with using the default gamma function (the "Mark Cohen" function) and its temporal derivative as basis functions to analyze event-related data in which stimulus presentation is staggered with respect to TR.

I find that this approach tends to "underfit" the actual data somewhat, and it appears that it could be improved if the "undershoot" could be included in the model. That is, the hemodynamic response dips below baseline several seconds after its peak, and then returns.

In SPM, a standard technique is to use the so-called canonical HRF, which is a combination of two gamma functions, the first modelling the increase and the second the undershoot. I would like to input this canonical HRF into waver and use it. However, I have been unable to find the exact equation for this function in the literature. Inspecting the SPM code, it appears that the basis functions (canonical plus one or two derivatives) are generated on the fly, and then "orthagonalized" in a second step, the necessity of which is beyond my understanding.

Does anyone know the formula for the canonical HRF and its derivatives, or a procedure to derive them? What about orthagonalizing them - is it worth worrying about?

-Jed Meltzer
Yale University