Jeff,

At 19:49 29.03.00 -0500, you wrote:

>I don't see a disagreement here; I didn't say why I would rather see real
>beads.  I agree that simulated data can be very useful, often preferable, in
>controlling the variables of a problem when you are trying to make precise
>quantitative determinations of how well a method is performing.  If such
>measures are available, I want to know them.  However, the proof is still in
>the puddin'--  I am not going to choose one method over another until I see
>how they operate on real objects (that we know something about, of course)
>on a real microscope.

That's the point. I saw wonderful deconvolutions of bead clusters, which
seemed almost perfect using software x (clearly better than the same beads
deconvolved with software y) , but deconvolutions of a real object (stained
actin fibers) gave rise to artificial fibers obviously not present in the
raw image (modification of algorithm x by the developer solved the problem).

The problem of recognizing possible artifacts is even worse when one
deconvolves very noisy images where raw structures are hardly visible. And,
in real life, one cannot check every 3D stack for artifacts...

> And yes, beads (real beads) aren't the best sample in
>all respects, as you have pointed out, but they are well-defined in terms of
>size and index of refraction, and they are convenient.

Therefore, one should not only look at artificial beads, but also at
artificial clusters of fibers oriented in different directions etc., as
Joachim Walter stated. (Of course there may be a problem finding the
respective 'real' objects).


Guenter



----------------------------------
Dr. Guenter Giese
MPI fuer Medizinische Forschung
Abt. Biomedizinische Optik
Jahnstr. 29
D-69120 Heidelberg
Phone (Germany or 0-)6221-486-320
Fax   (Germany or 0-)6221-486-325
e-mail: [log in to unmask]
------------------------------------------