How to automate the count of micronuclei in fluorescence images - a request for help

Dear Community,

We use someone else’s CellProfiler pipeline to count the nuclei of cells with a DAPI stain.
Then we proceed counting the micronuclei (small satellite bodies) by hand, calculate the ratio and perform some statistical analyses on the numbers. There must be a way to automate the second half of the procedure, too. I attached a typical example picture so that you know what i am talking about.
Can someone please let me know how they would go about this?


We have access to ImageJ/Fiji, CellProfiler, Imaris, and even Ilastik. I also feel comfortable executing and even (limited) editing code in e.g. R or C++ and we have the Keras library installed which would run on the CPU of an iMAC (no NVIDIA card installed).

Many thanks in advance!
I’ll let you know what worked for us.!

Edit: I just realized there are some posts on micronuclei already - all concerning CellProfiler. However, according this publication they miss > 8 % of these tiny speckles. And the posts are quite old, maybe someone developed a new/better approach since.

Hi @H2AX,

It looks a interesting problem. Though I am not sure about the publication you had shared,micronuclei.cpproj (680.4 KB) I tried with your image. The logic I tried is as follows,

  1. I segmented the Nuclei (normal big ones) using filter criteria.
  2. Convert them as a mask.
  3. I used the Enhance features to enhance the speckles
  4. I segmented the Enhanced image, but along with the micronuclei there are other regions which got segmented here.
  5. On this I applied the mask (of normal nuclei) which resulted only micronuclei.
    But this method missing one or two nuclei. But if you completely optimise with few more images it should work I guess.
    Please find the attached pipeline.
    Hope this helps!!

Fujifilm Wako Automation (Consultant)
For CellProfiler training or optimised pipeline write to,

Read more on our site.
Yokogawa CV8000 - The Ultimate in Confocal HCS

Join us at,

Dear Lakshmi,

Thank you very much for your input! In the meantime, i have optimised a pipeline which works pretty well on a bunch (75) of such images - usually +/- 1 object difference from the real count. The false positives are easily identifiable (by eye), as they tend to be of an elongated shape - maybe there is a measurement module for circularity in CP? Also, i haven’t tried any enhancements as you did. We’ll see if this improves it in the upcoming days. Please find attached my solution for the time being.

P.S. I just improved the pipeline to 80 % accuracy, filtering objects by FormFactor.
Micronuclei_Segmentation_Pro.cppipe (24.3 KB)


1 Like