> >
> > We also looked into the Windows-based MetaMorph/AutoDeblur system, since we
> > already had MetaMorph.  As some on this list have stated, MetaMorph is
> > powerful and user-friendly.  The main drawback with AutoDeblur is the
> > practical xy size limit of images, which is 256x256.  Anything larger has to
> > be processed in chunks, so I've heard.  I don't think this is the system to
> > go to for speed.
> >


Hi,

  I wanted to respond to an inaccuracy in the above response and offer
some advice on utilizing deconvolution.  AutoDeblur doesn't have a
limitation in the data size.  The limitation depends on the amount of
system RAM.  Now you can purchase relatively inexpensive systems with
512 MB of RAM.  The software can process data in 256x256 "chunks" or
smaller to get around problems associated with modest amounts
of RAM.  For instance, this allows computers having 40MB of RAM to be
used to perform deconvolution.  For large images the size of the chunks
(or subvolumes) used can be determined in a few ways.  The subvolume
operation can be disabled completely, the system can use the largest
possible subvolumes based on available system RAM, or a default subvolume
size can be used.

  For blind deconvolution, there can be an advantage to breaking an image
into smaller subvolumes and processing different parts of the field of
view separately.   This allows the deconvolution to account for varying
imaging characteristics accross (or deeper into) the field of view by
utilizing a different PSF for each subvolume.  Utilizing only one PSF for
an entire collected stack of optical sections does not allow for this.

There are a couple of other questions to ask as you start to investigate
deconvolution for your applications:

1) Am I trying to obtain intensity meaurements from the deconvolved result?

If Yes, than methods such as no-, nearest, and multi- neighbors
(deblurring methods) should be used with extreme caution.  These
methods are not intensity preserving and produce negative values
in the deconvolution result.  These numbers must be handled before
the result is viewed.  Inverse filtering methods typically suffer from
this same intensity loss problem depending on their implementation.

2) Am I going to obtain morphometrics from the deconvolved results?

If Yes,  again the deblurring and inverse filtering methods should be
used with the understanding that these methods cannot begin to restore (or
estimate) spatial frequencies which the micrscope system could not pass.
This is the "missing cone" region in the optical transfer function.   It
is the reason that straight vertical line intensities and flat sheets of
intensities cannot be imaged by a standard fluorescence microscope.  After
applying deconvolution methods like inverse filtering or deblurring
methods, these spatial frequencies remain absent from the result.   An
exmaple result of this can be seen for microspheres (assuming the sampling
along the optic axis is fine enough to avoid aliasing). Deconvolution
results obtained using inverse filtering or deblurring methods results in
bead reconstructions which appear "diamond" shaped when viewed in an
XZ or YZ planar view.  Conchello and Preza have a result demonstrating
this (for an inverse filtering technique) at their Washington University
website.  Iterative methods (ie. blind deconvolution or non-blind
deconvolution), can produce restored microsphere results which are more
sphere like as they attempt to restore spatial frequencies to the missing
cone region.

3)  Am I going to use the deconvolved results for publication or display?

If Yes, I usually will use the deconvolution method most appropriate for
making my point.  An example of this is when I am training someone on
collecting and deconvolving stack of fluoresecent optical sections.  I
will often collect the stack, perform some simple data correction on it,
deconvolve it using deblurring, inverse filtering, or blind deconvolution
at the fastest performance setting, and then generate a "spinning"
movie.   It's a case were I'm not going to be making any measurements.
If the fluorescent signal is relatively strong,  I will try a deblurring
method first for this purpose.



Best regards,
Nate


Nathan O'Connor
Vice President-Director of Engineering
AutoQuant Imaging Inc.
Watervliet NY 12189
ph: 518 276 2138
email: [log in to unmask]