Masking cells in 3D volume

Hi Guys (Mark),

I have a difficult case to segment, tips would be greatly appreciated!

Contained in the attached file are 5 axial images taken sequentially from a tif stack. My problem is that locally around each cell there is a uneven amount of noise. Sometimes the cell signal is weak and there is low background, sometimes the cell signal is high and there is high background (higher than the lower signal cells).

Any tips on how to best pre/ process these images for recognition?

Thanks!
   -J

Hi Josh,

I’ve taken a look at your images and I’m hard-pressedon two things:

  • Discerning where one cell ends and another begins (since I’m not familiar with the biology)

  • Do you want a cell segmentation for each slice or something for the volume as a whole?

  • I take it from your description that while the slice-to-slice foreground-to-background intensities may differ, do the intensities for each region also differ from slice-to-slice? That is, does the background intensity stay in the same dynamic range from slice to another (for example)?

In any case, I’m looking at the output of the MakeProjection module and I’m wondering whether the Power method might be useful, with the frequency set to something like 3 slices. It seems to capture some part of the light and dim portions and perhaps the identified objects can be used as a per-slice seed for IdentifySecondary? Just remember to save the projection as a .mat file if you look into it.

Regards,
-Mark

Since CP doesn’t have the ability to sort volumes what we’ve been doing is using it to find the cells and then generate a mask which I can pull into Matlab or Amira. Each slice is generally comparable from one to the next. The real problem is that different areas of the slice have different background intensities AND the intensity of each cell differs such that it could be below the intensity of background in another part of the image. The background you see is essentially the remnants of CT-like noise (shadows) that we can’t process out.

Your idea about using make projection looks interesting…