Can CellProfiler work with speckled grey backgrounds?

Dear CellProfiler community,

I am currently investigation the possibilities to automatically count a series of live cells images grown of a special coating. Although cell counting may appear as a simple problem to overcome using ImageJ or CellProfiler, I am running into specific problems. As I mentioned before, the cells were grown on a special coating. This makes sure the cells attach and grow as a mono-layer. However, this also results in a “grey speckled” background. Since the cells I am trying to count display the same aspect, I have a hard time differentiating cells from background.

As an example I have uploaded three example images:
http://imgur.com/mEPM1
http://imgur.com/JSWyY
http://imgur.com/QEDaP

These examples show the difference in morphology and density of the images. I’m trying the avoid having the count these cells manually, as you can probably understand.

Before I really dive into the depths of CellProfiler I ask you:

  • Is there a way to set CellProfiler to discard the speckled background?
  • Is there enough contrast between background and cells to accurately distinguish them?

Your feedback is very much appreciated!

[quote=“maas132”]
Before I really dive into the depths of CellProfiler I ask you:

  • Is there a way to set CellProfiler to discard the speckled background?
  • Is there enough contrast between background and cells to accurately distinguish them?[/quote]

You 2nd question is the one that’s really important. CellProfiler was originally developed and optimized for fluorescent images. Automated cell segmentation from transmitted light images is notoriously tricky, not just for us but also for those in the biological computer vision community in general. Unfortunately, it doesn’t seem like your images have enough contrast for our object detection modules to really work. Is it possible to use a fluorescence stain of some kind, where the contrast will be high between the cells and the background?

Best wishes,
-Mark

Hello Mark,

Thank you for your reply. I too realize that CellProfiler (and ImageJ by the way) are better at detecting cells that clearly differentiate from the background. Adding fluorescence would be ideal in the situation. However, we work with primary cells directly from the operating room. Adjusting the cells to express GFP is not an option to us, unfortunately!

I have already experimented editing the phase-contrast images by adding a blur, adjusting the threshold and then inverting the image. This results in a cellcount up to 25% from my manual count. 25% is still too much, but hopefully I will get closer to the manual counts after some more research.

Thank you for your feedback so far.