This looks like a good attempt at filtering by shape/eccentricity. One tip is to use DisplayDataOnImage o that you can see the individual objects’ measurement (e.g. eccentricity) values so you can intuit the best threshold. I added this to the pipeline attached, so that you can more precisely find the ecc threshold. Can you also use FilterObjects to filter based on size (Area)? You can add another filter criteria right in that same module, or add another module in series, which has the added benefit that you can see the results of each filter individually.
If you need further help, can you please annotate the images, labeling which cell is a DRG, and which channels’ info do you use to classify them?
P.S. I encountered a bug (which has since been fixed in the developer’s version, but not yet in the release version of CP), but might annoying to anyone else looking at your pipeline in the IdentifySecondaryObjects module. It’s also possible this misguided you too?
The details are here, but the workaround is to
(1) In IDSecondary, TEMPORARILY switch to the thresholding method = Otsu
(2) Change the Threshold Correction Factor from 2 to 1 (hit return, just so you know the pipeline registered this)
(3) In IDSecondary, switch back to the thresholding method = Automatic
Now the IDSecondary thresholding should be fixed.
DRG_Identify_DL.cppipe (12.3 KB)