Sorry it took awhile for the response, but brightfield images are notoriously much harder to analyze than fluorescent. See this FAQ. So my suggestions are, in order of most likely to least likely to succeed:
(1) Add a fluorescent marker! Or two, even better! DAPI plus phalloidin would make this trivial.
(2) Use another tool called ilastik (included with CellProfiler) to label the objects very simply, thus training a pixel-classifier in a very user-friendly manner. Then import the classifier into CellProfiler using the ClassifyPixels module, and segment the probability maps. An example that I recently provided is here: viewtopic.php?f=14&t=4098&p=12325&hilit=ilastik&sid=cb90268a6eb9901e0e3d7e1fcef89979#p12325
(3) Try CellProfiler alone. See the attached pipeline for my quick attempt. But I can’t say this method looks very promising.
DLproj.cppipe (6.93 KB)