Bob and all,

We are trying to smooth some data sets to a fwhm of 12.

We typically do this iteratively, with a combination of
3dmerge -1blur_fwhm
and
3dFWHM

runs, checking the output until we get near to 12.

Most of the time, this works fine.

However, some data sets, from certain centers (we are doing
multi-center fmri research) just won't smooth to 12 this way.

I would be glad to send along such a dataset -- we have many.

I wrote my own smoothing program, which does not use an
FFT approach, but which literally convolves the image with
a series of gaussian kernels. When I do this, I have no problem
smoothing the same data sets to 12 and beyond.

I suspect that it has something to do with the FFT approach.
The datasets that we have this problem with do appear to have
a somewhat more restricted dynamic range.

Here are the fwhm values I get as I smooth the same data
with 3dmerge -1blur_fwhm 10...15:

10:
sigmax = 3.90 FWHMx = 9.17
sigmay = 3.67 FWHMy = 8.64
sigmaz = 3.97 FWHMz = 9.34
:11:
sigmax = 3.91 FWHMx = 9.21
sigmay = 3.74 FWHMy = 8.81
sigmaz = 4.01 FWHMz = 9.45
:12:
sigmax = 3.94 FWHMx = 9.28
sigmay = 3.78 FWHMy = 8.90
sigmaz = 4.14 FWHMz = 9.74
:13:
sigmax = 3.95 FWHMx = 9.30
sigmay = 3.73 FWHMy = 8.79
sigmaz = 4.18 FWHMz = 9.85
14:
sigmax = 3.97 FWHMx = 9.35
sigmay = 3.76 FWHMy = 8.85
sigmaz = 4.27 FWHMz = 10.06
:15:
sigmax = 4.00 FWHMx = 9.42
sigmay = 3.82 FWHMy = 8.99
sigmaz = 4.30 FWHMz = 10.12

Any ideas?

Lee
: