I'm about to return to an old research interest, with (I hope) better and
more modern tools.
My goal with this posting is to find someone who is already doing this and
has ImageJ code to share. I will almost certainly need to modify it, but
a good baseline would be nice. I'm primarily interested in Java code.
Input: a stack of images (serial sections through retina tissue)
Output: 3D locations and sizes of spherical objects (cell nuclei) which
are stained a specific COLOR and nearly all the same SIZE.
(nuclei for RPE cells - mostly blue on a yellow background;
I'm also interested in characterizing the yellow background)
I suspect that I will eventually want a semi-automatic system, where the plugin
locates the easy ones and then allows a trained observer to edit (adding some that
are missed, removing some that are bogus, splitting "merged" objects, etc.
Image registration will also be a (minor) issue - that part looks fairly easy
(but I welcome pointers to the "best" solution - note that "best" for me will
involve Java code that I can freely modify)
The output (3D locations and sizes) will be fed to non-ImageJ programs, for Voronoi/Delaunay
based statistical analysis. In a (later, more advanced) version of this project, I *might*
move the Voronoi analysis into ImageJ and perhaps use analysis of the current VD to help
refine/edit the set of centers. But...that's on the horizon, right now.
The images will come from "whole mounts", with a motorized stage controlling x-y location
and focus determining Z. I know that there are ImageJ versions of software to do this and *might*
consider using them in the future - but for now I am constrained to accept a stack of images produced
by other software. Note that this makes the "image registration" problem pretty easy. However, there
is a higher level registration issue which involves finding landmarks at lower resolution to establish
an overall World Coordinate System. The preparation may also involve registering two pieces of tissue
that WERE bound together, but were then dissected apart. This registration will probably be based on blood vessels (which would be nice to do automatically) or on observer-identified fiduciary points.
So...who has stuff I can steal?
General replies should go to the mailing list - if you have something to share that works better one-on-one, feel free to contact me at <
[hidden email]>.
--
Kenneth Sloan
[hidden email]
Locked
