Hello Lukas:

As I see it, there are basically two operating modes for 3dDeconvolve:

1) Multiple linear regression -- Use program 3dDeconvolve to model the fMRI
signal as a linear sum of the input stimulus functions. In this case, you
might want to "preconvolve" the experimental input binary sequence(s) with
the assumed shape for the hemodynamic response (say, a gamma variate function),
using program waver. This is then used as the input stimulus function for
program 3dDeconvolve. (You might want to use the "-peak" option of waver, to
keep the amplitudes reasonble.) Also, in this case, you would normally set
the 3dDeconvolve maxlag = 0.

2) Deconvolution -- Use program 3dDeconvolve to estimate the hemodynamic
response (aka, IRF) for each of the input stimuli. In this case, the input
stimulus functions should usually be the "raw" binary sequences. That is,
the binary sequence(s) should NOT be preconvolved with the assumed hemodynamic
response, since you are attempting to estimate the hemodynamic response (IRF).
So, no waver. Also, you should set maxlag > 0.


The multiple linear regression approach is usually more appropriate for
block type designs. The deconvolution approach is usually more appropriate
for event related designs.

There are probably exceptions to all of the above statements.

Doug Ward