H&E Analysis: Nuclei Segmentation


Hello! Thank you for offering this awesome software to the public!
I am analyzing the cell shape and density of H&E stained tendon sections. I am able to separate the stains in some of my images with the UnmixColors module. Other images have fatty/scar tissue that does not allow me to unmix (example below). Because of this, I successfully isolated nuclei in two ways: by unmixing, and by thresholding for hematoxylin. I will analyze one of the two output images based on the section and the resulting image.
I am having a problem with segmenting the nuclei. I get both over and under segmentation, and I am unsure how to tackle this problem. This occurs mostly with the fatty samples, in which the UnmixColors isn’t used/effective. These samples are more cellular, with the nuclei clumped closer together.
What are the best strategies for optimizing segmentation? I have tried both the Morph module and the declumping techniques within the IdentifyPrimaryObjects module, but neither of these led to the desired results.
Thank you for your time!

H&E.cpproj (426.8 KB)


Hello there,

We’re very glad seeing you try multiple approaches and have reached certain success.
To be realistic, allow me to say that this kind of issue (nuclei clumped together in histology slide) is near the limiting edge of image segmentation. If it’s even difficult for human eyes to decide the separation, it’s not so fair to ask a software to perform beyond that. If later we’re aware of a better solution, we’ll gladly update to you and the community.

As for your particular image, I guess we can only hope to tune the segmentation to reach an “acceptable” detection, and/or include per-image quantification instead of focusing too much on cell-by-cell level.




Thanks very much for your quick response!

Yes, I agree this segmentation issue challenges even the limits of the eye. I just wanted to make sure I wasn’t missing any other tricks for segmentation optimization. As is, the pipeline is likely good enough to proceed with my analysis.

Thanks again! Your support is very appreciated.