Segmentation help

I was hoping someone might be able to help a little with segmentation on these images. Briefly this is an antiviral assay - cells were grown in 96 well plates and treated with different antiviral drugs prior to infection.
Cells were fixed, stained with DAPI / cell mask blue to outline nucleus / cytoplasm (both same fluoresence - I am more interested in being able to outline the whole cell) and then stained with antiviral antibody / alexa555 secondary. Goal of the assay is viral antibody expression between treatments in terms of the number of cells infected / intensity of staining.

Segementation of uninfected cells is not brilliant but ok (image 1). Unfortunately the infected cells fuse to produce large multinucleated cells which makes segmentation difficult (see image 2) and to make it more difficult the nuclei are saturated. Is there a way to segment these multinucliated cells so that each nuclei is counted, even if the cytoplasm allocated to that cell is not accurate - that would at least enable calculations of infected versus uninfected cells. Any other thoughts as to how quantify? Integrated intensity?

I have attached my pipeline. Any assistance greatly appreciated.

Thanks in advance
130707 final pipeline.cp (13.3 KB)

Hello Phil,

Sorry for the delay in answering, but my first suggestion before doing anything else would be to reduce the saturation in the images. In fact, there is an oddity to the intensities, such that it seems to be causing CellProfiler to segment badly. What I mean is, I looked at an Intensity histogram of the 1.tif DAPI image (attached), and the maximum intensity is actually just above 0.0625, or on an absolute scale, above 4096. Why this is significant is that it is quite common for microscopy cameras to be 12-bit, i.e. range from 1-to-2^12 = 1-to-4096. Your images appear to have a large peak at exactly 4096, but then there is some “spillover” into a range greater than this. I don’t know what is causing this, but regardless, this much saturation is not optimal for analysis. Plus there may be some other artifacts occurring when the CCD gets so saturated.
(I should also say that the file format is 16-bit, also common for tifs, so that this causes the images to only use the bottom range of intensities, from 0 to 2^12/2^16 = 0.0625. But again, the pixel values should be pinned at 0.0625)

CellProfiler makes some rescaling for display only, and in theory this shouldn’t matter, but I noticed some odd segmentation issues especially in the “declumping” steps of IDPrimary. If you turn off declumping, the segmentation looks reasonable. I suspect that since the threshold found by the module is so close to the absolute threshold (0.0625) that the declumping algorithms might be behaving oddly. I could debug further, but perhaps you have some better, less saturated, images by now?