Counting rare events in brightfield images


My lab typically uses brightfield images (admittedly not what CP was optimized for) but I’ve had success particularly through using the UnmixColors module. I’m currently having some issues with analyzing images in which the targets to be quantified are small and fairly rare (often under 10 per image). When I try to select these objects, CP always overselects as if it’s expecting more objects. Any recommendations?

Images linked; features of interest are the bright pink/magenta punctate signals against the hematoxylin counterstain.


I put together a pipeline which should help. I kept the UnmixColors idea, but only for detecting nuclei (which you didn’t even say that you needed anyway). I’ve found that UnmixColors can be tricky for isolating some stains, and you often need to find the closest match in color, regardless of the actual name. Also, some creative color arithmetic seemed to do a good job at highlighting the punctate features.

Forum attachments aren’t working right now, so here’s a Dropbox link:



Thanks much for your help. Since posting this yesterday, I got lucky with one of the color combinations (PAS for the red and hematoxylin for, well, the hematoxylin). I got much better segmentation through this, but your pipeline looks very robust. I appreciate assistance and the fantastic program your team has developed.


Great to hear! There’s often more than one way to skin a cat :smiley: