H&E Cell Counting Malfunction

Happy summer!

Thank you for all of your work in developing this amazing software and for providing help to us through this forum.

I’m currently having an issue with using a pipeline to count the number of cardiac cells in a histological H&E image (and through doing that, find the cell density in that image). I’ve been able to develop/modify a pipeline accomplishing what I desired. Basically, the pipeline counts the number of nuclei in the image, and I use that to represent the number of cells. However, I’ve tried running the pipeline on 20+ different images; and while it works well on some images, on others, it is unable to count any nuclei. I’m not sure what is wrong with the pipeline; I’ve tried adjusting threshold values, diameter size, and everything, but its effectiveness is very variable.

I’ve attached the project with the pipeline and three images. The image, labeled with Rat 81, works well with the pipeline. The two images, labeled with Rat 90, are examples of the problem I encounter (from a nuclei count that makes little sense or no nuclei count at all).

Is the issue with the quality of my images? Or is there something that I need to change with the pipeline to make it work better?

Once again, thank you for your time and consideration.

CellDensityHL(working).cpproj (157 KB)

Hello Ha,

Sorry for the delay. Glad you are trying CellProfiler and you have gotten quite a lot configured so far! Histological image analysis is notoriously hard because the samples vary a lot, at least to the computer.

First off, your NamesAndTypes is configured in a way that it was not designed for. Rather than specify a separate category for each(?) image, it is intended to find ALL images of a certain type (usually, channel/wavelength). So in your case, if all Rat samples are to be analyzed the same, then you can simply have a single category that finds the text “Rat” in the filename (or any consistent text across all images). Otherwise, all downstream modules would not process all the images, or you would have to duplicate all the modules many times.

I adjusted a number of parameters in a new IdentifyPrimaryObjects. Take a look at the attached pipeline. I disabled yours, but kept it for comparison. I think it helps, but it is hard to have a very robust analysis pipeline for histological images.

DL_CellDensityHL_working.cppipe (14.3 KB)