Recently Dave Beebe wrote:
 
>OK, so confocal and deconvolution methodologies have their place
>for achieving the ultimate in resolution when imaging certain kinds of specimens.
>But what about in real time on an everyday basis when you are sitting
>at the 'scope scanning through a new specimen just trying to get a sense of what's
>going on.  Is deconvolution useful for these situations, or must I wait for the
>computer to digest and regurgitate my image before I know whether I have
>something interesting?  In more direct terms, is deconvolution practical for rapidly
.>and easily evaluating new specimens where the major purpose is getting
> few informative pix, not the ultimate in 3D resolution?
 
and  Aryeh Weiss wrote:
 
>I think it would be useful if users of wide field deconvolution
>systems post some processing times for "typical" dataset,
>specifying the computer configuration (hardware and software)
>which they use
 
At the risk of restarting the deconvolution thread again, I would like to
respond to these two points.
 
VayTek's goal from the beginning has been to develop a practical
deconvolution "toolbox".  The toolbox approach means a complete
integrated system with several algorithms, each appropriate for
various situations and applications.
 
We know it is important to have rapid feedback in real-time situations.
To do this with deconvolution means integrating the data acquisition
system with a rapid deconvolution algorithm.  We have done this on
both the Macintosh and the PC.
 
With our system, the user can easily move the stage, view a live image,
capture a single image, and apply a single image nearest neighbor
deconvolution.  This process takes about 5 to 10 seconds.  Color images
(which cannot be captured on a confocal system) take a few seconds longer.
 
If the user wants a slightly better deconvolved image, he/she can use
a nearest neighbor algorithm with three consecutive optical slices.
It takes only slightly longer to capture the three images.
 
There will be objections by others about the value of nearest
neighbor vs. constrained iterative and that this approach is not valid.
(Or perhaps they have a vested interest in pushing the constrained
iterative.)  However, we know from experience, (many of our users will
verify this),  that usually there are small difference in images deconvolved
with nearest neighbor and constrained iterative.  The nearest neighbor
algorithm is very useful for determining what the image will look like
when deconvolved with the constrained iterative - and doing it
quickly.
 
After a specimen has been previewed with the nearest neighbor and
the proper settings have been determined, a full stack can be captured
and deconvolved with the constrained iterative.
 
In other words, the answer is yes, there is a way to use deconvolution
in a practical, real time situation.  It may not be as fast as the confocal,
but it is pretty close.  There are trade-offs however, one being flexibility.
 
Our integrated system does make a good "first line" system for some
researchers with some specimens.
 
Regarding typical times:
 
This a little more complex since there are several variables involved.  But I'll try.
 
In general, the times for Macintosh and PC are pretty comparable.  All these times are for VayTek's imaging systems.
 
1)  512 x 512 grayscale image with a DSP board - never more than 4 seconds/image using nearest neighbor - any size Psf
 
2)  512 x 512 grayscale image without DSP board - depends on CPU speed - for Pentium 150 - no DSP board - 2 to 10 seconds - nearest neighbor - with a small PSF;  longer with large PSF
 
3)  512 x 512 color image with DSP board - 16 Mb memory - nearest neighbor - 10 to 15 seconds
 
4)  stack of 25 512 x 512 grayscale images - with DSP board - sufficient memory - small PSF - optimal number of iterations - constrained iterative - 5 to 15 minutes
 
5)  Data acquisition - live image - real time;  capture and store a single 512 x 512 image to disk - 1 to 2 seconds
 
6)  Time to capture a single image and deconvolve using a single image nearest neighbor algorithm - 5 to 10 seconds
 
7)  Time to capture a stack of 25 images - 512 x 512 grayscale and write to disk - 25 to 30 seconds
 
8 )  Time to deconvolve a stack of 25 images 512 x 512 grayscale using nearest neighbor and DSP board - about 2 minutes
 
I hope this answers your questions