microscope image processing

mangalam at SALK-SC2.SDSC.EDU mangalam at SALK-SC2.SDSC.EDU
Fri Sep 4 16:35:54 EST 1992

Regarding your request for information regarding the capture and processing
of video images from a microscope at 1000X1000X12-16 bits, for an
application running on a *nix workstation.

>The application should meet these requirements
>(or at least part of them) :
>- deliver a good and fast video display of the image(s)* (see below)
>- standard operations on images (+, x, -, convoluation,*
>  noise reduction, linear or non linear LUTs)*(with the free NCSA PALedit)
>- counting cells*
>- numbering nuclei as objects (contours, areas, ...)* (I think)
>- integrating the signal within these objects*
>- define masks and apply them to other images*
>- show the data in statistical format (histograms,
>  correlations, ...) or export it in a format compatible with  
>  standard statistics processing applications (Macintosh
>  or MS/DOS Excel,Cricket Graph, or Unix based applications,..)*
>- print the images to a thermal/laser printer *
>- drive the microscope stage (z axis, or even 3 axes)
>- perform 3-D reconstruction and display from a user
>  defined angle. *(sort of)
   There are number of answers, none of which (of course) answers all your
requirements.  Surprisingly, many of them can be met by Wayne Rasband's
(free) NIH Image running on a Mac.  It can perform all the points marked by
a * above, including direct support for at least 2 video frame grabbers. 
The major limitation is that it was not built to handle the _huge_ image
files that you mention and it has the unfortunate limitation that it will
only handle 8bits of color/greyscale.  If you can reduce the amount of
data, or you are willing to write the modifications to his routines (Pascal
source code is freely available)), then you could save a great deal of
money. (If you were to re-write the source code you could also link in the
routines to use a very fast coprocessor like the Tuplex i860 for the
graphic convolutions/transformations.) The first thing to do is to download
the documentation and the program to see if it is worth the effort. Image
is available from alw.nih.gov in /pub/image.
   If you absolutely demand the size of files you state, only a _very_ fast
workstation will be able to handle them in a 'real-timely' manner, an SGI
Indigo R4000/Elan at minimum and up. SGI recently released it's Reality
Engine - 8 i860 superscalar processors dedicated to graphics processing at
(only!) $60,000 academic on top of whatever VME box you put it in (AMAZING
   The SGI recently released a set of video manipulation tools for the
Indigo - it requires a $2500 board (that only fits into the R3000 based
Indigo because of the size of the R4000 board) but some software tools are
included in the system software. I rather doubt that this board will handle
1Kx1Kx16 bit video capture, though - ask SGI about it.  
   At any rate, to do the analysis, you will have to use some additional
   After digitization, you can reconstruct the image with a program called
SYNU (SYnthetic UNiverse) from UCSD and the San Diego Supercomputer Center
(costs ~$100 to cover costs) which can handle huge data sets and render
beautiful reconstructions (call David Hessler @ 619 534 7968 or Steve Young
@ 619 534 3539 - unfortunately, it only runs on SGI hardware)
   There is also a horrendously expensive (~$20,000), but quite capable 3D
volume rendering/analysis package called VoxelView (Vital Images; phone:
515-472-7726 Ext 118, fax: 515-472-1661, email: userserv at vitalimages.com)
which can do some of the other kinds of image analysis that you require
(also requires SGI hardware).
   Summary - try NIH Image first, rewriting that code that you need to,
then use SYNU if you need it for better reconstruction.

Std Disclaimers Apply...



Harry Mangalam                                   Vox:(619) 453-4100, x250
Dept of Biocomputing                                   Fax:(619) 552-1546
The Salk Institute                             mangalam at salk-sc2.sdsc.edu
10010 N Torrey Pines Rd                        mangalam at salk-sgi.sdsc.edu
La Jolla CA 92037                                    mangalam at salk.bitnet

More information about the Bio-soft mailing list