> Why is the shape look different from the pictures in the
> AFNI 3dDeconvolve manual? Is this because my
> experiment is an event related? Or because the
> deconvolution was runned with unmodeled signals?
> Otherwise, something wrong?

This seems to be the trade-off of modeling brain activities with different approaches between assumed of IRF shape such as gamma variate and deconvolution.

Deconvolution does not make any assumption about the shape and duration of the hemodynamic response function. Thus the modeling is totally based on whatever available from the data. The advantage is its sensitivity and flexibility, but the downside is that it could inflate the detection and mispresent the impulse response function with noise.

On the hand, the assumption of a fixed IRF shape reduce the chance of mixing noise with signal, but it is less sensitive and also loses the flexibility on modeling the shape and duration across different voxels.

Some people argue using a set of basis functions might be a compromise between the above two approaches. If you want, you can try the -stim_times option in 3dDeconvolve:

http://afni.nimh.nih.gov/afni/doc/misc/3dDeconvolveSummer2004/

Gang