There is a similar question here which may provide some useful answers:

https://www.researchgate.net/post/Can_anyone_help_me_find_a_method_to_semi-automate_the_analysis_of_mitochondrial_morphology_with_electron_microscopy_images

In general, this segmentation task appears to be not trivial.

Cheers,

Jochen

Thanks, Jochen.

The top answer suggests a supervised classification approach (Weka-Fiji). Although I have played around with Weka a long time ago, it was not for segmenting nuclei. I will try it and see how it goes!

Could you post a (much) higher resolution image?

I thought it would be useful to use the Blue channel. I attached Blue minus Green; this beautiful piece of art is mainly jpeg artifacts of the original image!

Apologies! Here's a better (the best I could get) version of the image.

PS: Are .tif images not allowed as attachments here? I tried and got an error message - "The file type you tried to upload is not allowed. Please convert to a different type.".

Girish, excuse my ignorance, but are you trying to delineate the number of IHC positive nuclei (seen in the higher resolution image) from the negative and come up with some sort of analysis? What is the thickness of your sections?

What software do you have available to you?

Hi Ronald,

I work with researchers who work mainly in the protein biomarker discovery area. So, usually, analyses of IHC-stained images consist of detecting both IHC positive and negative nuclei and calculating the proportion of positive nuclei. This, along with the IHC intensity is what they generally use to perform further analyses.

I am not a biologist and thus, have very limited knowledge of how the actual sections are prepared (they are prepared by colleagues who work in the wet lab).

As for software, I have Aperio's IHC Nuclear algorithm available (which seems to have three segmentation methodologies rolled into one), and also have an algorithm developed in-house. These work good enough for some cases. In cases like the image I've posted, the detection of the haematoxylin-stained nuclei is poor. This is expected, due to the heterogenous staining of the nuclei. It is among the extreme cases I've seen. I also have Fiji (ImageJ) available, although I rarely use it.

My interest is in learning about computational techniques which could possibly be used to better segment such unevenly-stained nuclei.

Girish, contact me off-line and I may be able to put you in touch with one of our computer gurus?

[email protected]

I gave it a try with the better resolution image... and I did not succeed. So -for now- I agree with Jochen that segmenting those nuclei is not trivial!