IdentifySecondary (and CP in general) assumes fluorescent images, not brightfield, so I'm not surprised you didn't get a good outcome here.
For transmitted light images we often suggest a workflow that combines ilastik and CellProfiler, and I definitely think that would work for you. Because your TL images are so textured though, you might be able to get around that by using the EnhanceOrSuppressFeatures->Texture module on your TL images; that'll turn them into a psuedo-fluorescent image that would act as a good substrate for IDSecondary. Here's a zoom in from the workspace viewer showing some cells where I identified the nuclei in IDPrimary and used that Enhance image for IDSecondary.
EnhanceTL.cppipe (8.4 KB)