Identifying nuclei

Hi,

This is usually a straight forward part of the pipeline for my analysis but I am starting to struggle with identifying the nuclei in my images and I am not sure what the reason for the change is. I have attached my pipeline and an example screenshot to illustrate the problem. It seems to be over segmenting the nuclei and over estimating their size but despite altering settings such as clumped object identification etc., the problem persists. Any help with the settings to improve the nuclei identification would be greatly appreciated,

Many thanks,

Barry
example.cp (7.94 KB)


Hi Barry,

I suspect that if you right-click on the top-left image and select Image Contrast > Log normalized, you will see that the cells themselves have some non-specific staining. If so, this mid-level intensity (dimmer than the nuclei but brighter than background) is probably what the thresholding is detecting.

You may want to consider either increasing the threshold correction factor or switching to a different thresholding method. In your case, I would suggest trying Otsu Global and 3-class thresholding with the middle class set to background. This will cause the middle intensity levels to be counted as background.

As a sidenote, you usually do not want to apply an illumination correction function from one stain with an image from another stain. The two may look similar but are not really comparable, so an illumination function should be generated for the phalloidin channel separately.

Regards,
-Mark

Many thanks Mark,

We’re using Hoechst staining for the nuclei and have noticed some shadowing around that area despite titrating down the amount of stain used. Its probably this shadow thats being picked up. Good tip about the illumination correction also, I was under the impression that the same correction should be applied to all channels but have modified the pipeline accordingly.

Best wishes,

Barry