Help counting total cell numbers

Hi Everyone,

I was wondering if all the cellprofiler experts in the forums might be able to help our research group with some automated cell counting. Our group studies optic nerve regeneration here in Boston and we have been trying for some time to streamline our cell counting efforts, which at this point takes up a great deal of time. We are interested in counting the number of cell bodies that are stained, the difficulty we have had is that the number of axons makes basic automated cell counting a problem. For example, in the attached image we are interested in the total number of green and red cell bodies. but as you can see the green axons obfuscate the image. In your collective opinion, do you think cellprofiler is capable of analyzing such an image for cell bodies?

Thanks so much!


Hi,

I’m attaching a pipeline that should get you a little closer to a solution (I hope). The key features are the following:

  • Use a special filter to try to highlight regions that look like dimmer regions surrounded by a brighter ring (which I assume are the bulk of the green-stained cells).
  • Perform further filtering based on morphology to isolate real cells.

Regards,
-Mark
2013_09_26.cp (12.6 KB)

Hi,

Mark provided as good a solution as could be expected given this tough problem, but reading your query I had a couple questions too:
(1) We are assuming you want independent counts of green and red cells. Is that correct? Or do you want simply those co-localized?
(2) Assuming you do want to independently count green-only soma, how reliably can you count them by eye? I find it pretty difficult! As long as you don’t have too-high expectations that CP will outdo your eye’s performance, this might work.
(3) If these are axons, have you tried other markers like MAP2 that don’t tend to stain axons? (Apologies if this was too obvious! But I thought I’d ask)

Another tack would be to pre-process with ilastik (a pixel-based classifier) and then use CP to segment the output probability map (hopefully higher probability near green soma). You would simply draw on the green cell bodies and let ilastik’s machine learning classifier highlight those regions that it thinks are green soma-like.

Cheers,
David