Dear Aser,
When going with the default settings in SPM a boxcar function is created based on the condition onsets and durations, which is then convolved with the canonical HRF, no matter whether onsets are close in time, "active" windows overlap or not. This might be different in other programs, maybe the model settings are adjusted automatically based on input.
In SPM, the closest and most straight forward implementation of a model similar to the AFNI "deconvolution" (for the difference between this deconvolution and the deconvolution from PPI analyses in SPM see slide 30 in https://afni.nimh.nih.gov/pub/dist/edu/latest/afni_handouts/afni07_advanced.pdf , for the details of AFNI implementation see http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.454.5661&rep=rep1&type=pdf ) should be a Finite Impulse Response model = FIR model. It does not assume a predefined shape of the elicited BOLD response, but rather you get a series of stick (also impulse response) functions, with the different regressors modeling the first, second, ... nth time window after stimulus onsets. Based on the estimates you can then plot a time course of the "activation" within a voxel or in regions. Instead of stick functions one could also think of e.g. tent functions like those available in AFNI or a Fourier Set (like in SPM) or other functions. In the end, the point is to have a number of regressors (which allows for flexibility) that can account for signal changes within an expected time window.
Best
Helmut
|