I suspect that this is fairly easy - but I seek guidance from more experienced developers:
*Gray-scale stack (on the order of 100 slices) of moderate resolution (call it 1024x512x128).
*objects appear as dark, mostly circular areas (boundary is "implied" by a surrounding ring of small bright blobs)
*probably too difficult to automatically segment in the time allowed
*easily sketched by a trained observer
*may require changing viewing resolution and moving back and forth in the stack for verification
*objects are branching structures, but VERY big in cross-section and VERY simple branching. Assume that
polygons from the same object overlap in x-y and polygons overlapping in x-y ARE from the same object.
*multiple objects in each slice - I *think* I prefer to gather information slice-by-slice, rather than
object-by-object, but I'm not fanatic about that.
What I think I want is merely a collection of 2D polygons, each tagged with a z-coordinate. Output in a form
suitable for further analysis. No need to combine polygons into objects. I can do everything else "offline".
That's the simple version. Being more ambitious:
*extrude each polygon to a solid (width = inter-slice spacing - the volume is close to isotropic)
*form (3D) connected components
*display in color, embedded in the gray-scale stack
*compute basic properties such as volume
*smooth the object (not simply a pile of blocks), display appropriately, and compute surface area (and a new volume)
My first priority is a data-gathering tool to provide computational support for the trained observer.
The bells and whistles come later. I suspect the final version will export the ROI's from the data-gathering program and a second program will want to read in the Stack and the ROIs and do the visualization/analysis. But, the final
version may well present the 3D reconstruction while data is gathered (but…not the version that will produce data
we need in the next month).
So…can I get this off-the-shelf?
If not, what specific tools/techniques would you recommend. Assume that my general preference is full-Java plugins.
--
Kenneth Sloan
[hidden email]
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html
Locked
