Cell Count H&E stain

Hi!

I’m recently trying to use CellProfiler to count cells on histological images (hematoxilin&eosine stain)of bone graft on rats with and without MSC.
It has been very dificult to create a pipeline to do the count. I already tried to analize this images on ImageJ using the ITCN plugin, but at first I coundn’t do it, because the program didn’t recognize between osteblast, graft, bone, blood cells. So I used Photoshop to isolate the nuclei of osteoblast firts and then make the count, with very successfull results (compared with manual count). But this photshop process is very hard and sensitive to human error, so the calibration took a long time.
My question is if can CellProfiler make the count and isolate the cells of my interest (osteoblast, osteoclast and fibroblast) on the original images without the photoshop process?, or if it doesn’t, how would be the pipeline to do the count on the images processed with photoshop?.

I attached an example of the original image and the same image processed on photoshop. The manual count of this image is 556.
I’ll be very gratefull for your help.
Thank you!



Hi,

Please first note that this is similar to other H&E images, like this one. The advice in that thread, especially my advice in the second post there is applicable here.

I attached a pipeline that you can start with. It really depends though on what you consider individual objects, which isn’t clear from the mask you provided (thanks for providing that version of ‘ground truth’ though!).
See my comments in the previous thread on how to optimize this.

Basically,
(1) Load original histo Image
(2) UnmixColors – used especially for histology. Choose H&E (or any other unmixing deconvolution that works)
(3) EnhanceOrSuppressFeatures (Speckles) – good for removing background, non-specific staining. "Speckles’ is kind of a misnomer here, but it helps nonetheless.
(4) IdentifyPrimaryObjects - Segments the cells. I am using 3-class Otsu, since even after enhancement, there are (at least) three intensity region types: (1) no tissue, (2) non-specific, dim stained, and (3) stained cells. I lump the middle class with the true background and am left (hopefully) with only the stained cells.

Hope that helps!
David
Pipeline_H_and_E_DL.cp (3.89 KB)