Hi, I work in a neuroscience lab acquiring 3D pictures of neurons at high
magnification.  We are using deconvolution on an NT machine.  Our images
have voxel size of 0.09 x 0.09 x 0.25 microns.

We have been using Autodeblur from Autoquant on an NT machine.  In our
experience, it is the best deconvolution software on the market not only
for PCs but for any platform.  The other available software require
empirically acquired point-spread functions which are very demanding to
acquire.  If the empirical point-spread function is not acquired under
absolutely pristine conditions, the deconvolution will impair your image
more than it will improve it.

For this reason, my lab actually invested in a PC workstation just to be
able to run Autodeblur.  We bought a dual-processor Pentium III 733 MHz
with full speed cache.  It still takes about 8 hours to deconvolve a file
of about 300x300x60 pixels.  However, the improvement in resolution is
considerable.

If you have a multiprocessor SGI machine you might be able to process the
same image in about 15 minutes.  However, there is no currently available
software for the SGI that does not require empirically-acquired
point-spread functions. And this means that in exchange for your fast
processing time, you are committing yourself to months of headaches
learning about point-spread functions.

Of course knowing about point-spread functions is probably wise in the
long run, and struggling to acquire a good one is good training in
microscopy.  However, it will be very hard to pass on that training to
anyone who is not intereted in microscopy per se, but only wants to
produce data.

Wes Wallace
Dept of Neuroscience
Brown University