Histological image analysis is tricky. I made an attempt (see attached pipeline), and it seems to work ok for the test_1.png but not so well for test_2.png. Unfortunately, I don’t think this is easily doable without a much more complicated analysis. You’ll note in my pipeline that I tried a few other approaches (Enhance the nuclei, then use Morph on them to try and “fill in” the nuclei which tend to be doughnut-shaped, then segment), but none worked to my satisfaction.
In your pipeline, you weren’t actually using the UnmixColors output in the downstream modules. The output of UnmixColors are grayscale images, and so you don’t need the ColorToGray module either.
You could also try other approaches, like using ilastik, a pixel-based machine learning tool to train the program on nuclei and eosin areas, but I suspect that won’t help all that much.
The best thing you could do is try and get the H&E to be more penetrant into the nuclei (not sure how feasible that is) or to use a fluorescent nuclear marker (the suggestion that would help the most).
Hope that helps,
DLogan_HE_nuclei.cppipe (10.8 KB)