Identifying paxillin spots



Hi there,

I am trying to identify focal adhesions, but my pipeline identifies way to many spots.
I have tried everything I could think of with help from the forum, but my problem is not solved yet. I think the main issue is that the contrast between my focal adhesion stain (paxillin) and the background is not really good, but still by eye I can distinguish the spots from the background.
I attached my project with an image that gives me a lot of trouble. Can anyone help me?
PAXtest(2018-03-04)3.cpproj (100.2 KB)

Best Hanneke


Hi Hanneke,

Can you please upload the image as well? I Only see your cpproj.



Hi vchernys,

Thanks for your response, I had thought the image would be in the project file.
I had some trouble uploading it as a .tif, so now it’s a zip file: E2 014 (1.8 MB)

Best Hanneke


You had a good approach in using EnhanceOrSuppressFeatures. It really comes to adjusting feature size in that module and IdentifyPrimaryObjects parameters.

E2 014 (466.4 KB)

PAXtest(2018-03-04) VC.cpproj (403 KB)


Thanks for your reply again vchernys!
This pipeline works very well for the Green channel, but unfortunately not so good for the Blue channel, which is actually the paxillin I am interested in (sorry for not mentioning this before).
There are really few spots in these cells, for the image I attached there seem to be approximately 15 paxillin spots.

Do you think it is even feasible to extract these from the image?

And just because I like to understand things: Why did you use an RGB image and the ColorToGray module instead of 32-bit tif with selecting channels from the metadata as in the pipeline I posted?

Best Hanneke


And here is a file with only the paxillin channel in case it is useful:
E2 014 SUM (449.4 KB)


Hello Hanver,

Thanks for the clarification. It is good to use the metadata like you mentioned and it was really just my choice to use RGB. I was able to extract the channels from RGB by using the ColorToGray modules. Your approach works good too and might actually be good if you already have channels separated prior to CellProfiler analysis.
Looking at your image it should be possible to do it by the intensity difference. They do look about the same size as other dots so it may be difficult to filter purely by size.
Please take a look at this version. Looks like robust background might not be too bad of threshold option to use.
PAXtest(2018-03-04) VC2.cpproj (412.6 KB)