Trying to segment dim cells

cellprofiler

#1

Hello,

I been trying for a while now to segment some cells that are quite dim due to a depolymeriser treatment. (These cells are part of an assay that will include a range of compounds including depolymerisers and stabilizers). IdentifyPrimAutomatic works fine on the nuclear stain … but I have difficulty in getting a good (+consistent) result with IdentifySecondary. I have experimented with many settings. Typically some cells in the image might be segmented well, but then others have greatly enhanced boundaries.

So I was wondering if I might be able to get any further advice or suggestions. Any help or comments would be most welcome. I have attached some sample images (each with a nuclear and secondary stain). Note that although the images are dim, the cell bodies do become visually apparent if the contrast is increased greatly.

PS I have tried all the different algorithms, and experimented with threshold correction factor, lower/upper bounds, adaptive vs. global etc., and with “per object” methods. Also the images have already been illumination corrected.

Thank you (+ thanks also for making cellprofiler available to us),

Ernest







#2

Hi Ernest,

I am not having much success on segmenting your images using IdentifySecondary either; in general, you will have such problems if the stain you are using to detect an object is one of the variables you are changing at the same time. You will normally want to such a stain to be stable regardless of treatment (such as a membrane stain rather than a cytoskeletal stain for example).

I was able to have some success with using IDPrimAuto on the cell stain image (since there are more settings to adjust to get a good threshold), then convert the objects to an image, and finally using IdentifySecondary on this new image. I’ve attached the pipeline, but you’ll need to continue adjusting the settings to see if it meets your needs.

Regards,
-Mark
2009_07_27_PIPE.mat (1.37 KB)


#3

Hello Mark,
Thanks for the reply and the help.
It’s an interesting idea to use DPrimAuto on the Cy3 stain images … I will try it out (and the pipeline you made), to see if I can get a better result. I agree that one of the difficulties with this assay is that we do not have a cytoplasm stain, and need to use one of the targetted structures (the microtubule filaments) for segmentation.
Best,
Ernest