Nuclei 3D Segmentation


I am trying to segment nuclei within the Arabidopsis shoot apical meristem to quantify their volume following the section “3D segmentation” from the MGX manual, and I am having some problems. In particular, after using “ITK Watershed Auto Seeded” with different values (level) only a few nuclei are segmented and show weird shapes that not correspond to the actual nuclei.

Is there any possibility of manual 3D segmentation? As there is for 2D segmentation… Or should I play with other tools (which ones) before using “ITK Watershed Auto Seeded”?

Just in case, I write the sequence of tools that I am using:

  1. Brighten darken (2.0)
  2. Gaussian blur (x,y,z sigma = 0.7)
  3. Binarize stack (threshold 10,000)
  4. Hand-correct stack using Pixel Editor tool
  5. ITK Watershed Auto Seeded (level values tried = 500-1500)

Thank you very much in advance.

Are you trying to segment a stack with a nuclear marker? or a nuclear envelope (pore) marker?

I am using a nuclear marker. Well, it is a transcription factor that
it´s expressed in a particular domain within the meristem. I imaged
the samples at 8 bits instead of 12 or 16, so I don’t know it this
could be the problem. But I segmented the cells (not the nuclei) of
other samples imaged at 8 bits and it wasn’t a problem.

“Richard Smith via Forum” escribió:

It is kind of tricky to tell the nuclear size from a nuclear marker, if you turn the gain up or down, the nucleus will appear to get bigger or smaller. Same as you go deeper into the tissue, the nuclei will appear smaller. A nuclear envelope marker is better for this purpose.

For the data you have, you could do steps 1 & 2 and then use the Mesh/Creation/Marching Cubes Surface process with the same threshold. You could also run it after step 4. The output of this should be a mesh composed of balls, which you can then label with the process Mesh/Segmentation/Relabel 3D Cells.

Yes, you are right. I´ll try to get this kind of reporter to do it properly.

At least I tried your suggestions and the segmentation worked
(although the nuclei shape does not correspond to the real nuclei
volume because of what you said).

Thank you very much.



“Richard Smith via Forum” escribió:

you can use DAPI and PI staining, very precisely adjust scanning (Scanning quality/parameters is 90% success of your analysis). What you can do in the current imnage, is to comare size of different nuclei in the frame of coordinate system and cell types detection. Look out on HEPPP description.