Looks to me like you have some options:
- identify the white spaces (IdentifyPrimAutomatic), then you can make measurements to characterize them (i.e., measure the intensity of the brown/black within each white space using MeasureObjectIntensity). Looks like some of the cells are out of focus and don’t have white space but rather appear pretty blue, matching their surroundings - is it ok to miss those?
- identify the brown/blue nuclei (IdentifyPrimAutomatic) then identify the edges of the lacunae (IdentifySecondary).
- identify the white spaces (IdentifyPrimAutomatic) as in (1) AND identify the brown/blue nuclei (with a second IdentifyPrimAutomatic module) as in (2), then use the Relate module to see which white spaces contain nuclei and which do not. There is an example pipeline on the cellprofiler.org website that uses the Relate module (Speckles) so you can see how that works.
The choice will come down to which objects are most reliably identified. If you want to catch cells that do NOT contain nuclei at all, then of course #1 or 3 is best. The key for 1-3 is to separate the three-color original image into the color combination that gives the best contrast to see the objects you are trying to identify. Could you play around a bit with ColorToGray, trying different combinations, to see which might be best, then post the results here? You could also try InvertForPrinting, because that does a dramatic color conversion. If none of these produces a nice-contrast image where the objects of interest reliably stand out above the background, you might also try the FindEdges module. It will help make any non-smooth areas of the image stand out above background.
Once you identify the ‘objects’, one way or another, then you probably know you can use all of the Measure modules to measure all sorts of features. You can also use CellProfiler Analyst to do supervised machine learning classification based on multiple features to classify each cell as one of the different cell types, if a single feature is not sufficient to distinguish cell types.
Note also that if the big white space interferes with the analysis, like in the top of the image you attached, you can always have the first step of analysis be an IdentifyPrimAutomatic that is set to find very large objects, then Crop to exclude those big white spaces from analysis further downstream.
Keep us posted!