Cell Counting Using Cell Wall Outlines

Image Analysis
#1

Hey!

I’m trying to count the number of cells in my image (taken with ZEN). They’re stained for E-Cad, so it’s mainly showing the outlines of the cell boundaries. When I try to use IdentifyPrimaryObjects, CP doesn’t seem to know how to figure out what are my cells and what aren’t. Most of my cells appear to be thin rectangles, but CP keeps making more circular objects.

Here’s a pipeline I tried to use:
Cell Body.cpproj (646.9 KB) How can I create a pipeline that better identifies my cells?

Here are some examples of my images: Experiment-14-23 (dragged)-cad.tiff (1.0 MB) Experiment-10-23 (dragged)-cad.tiff (1.0 MB)

Any help would be appreciated :slight_smile: Thanks!

#2

Without Hoechst or some other nuclear marker, it will be very difficult to tell where one cell ends and another begins.

Most programs create the cell by expanding outwards from a nuclear objects. I don’t know that that is the case in cell profiler, but it seems like a good guess if your objects appear too rounded.

Here, you could probably use “low red” intensity, with a size limit to get the center of your cells, but to tell where the cell ends would require expanding along the red gradient. If it is an option, I would try a very small cytoplasmic expansion.

<== Doesn’t know cell profiler.
Even with a quick pixel classifier, it wasn’t easy to find where the nuclei would be.

image
image1331×1337 1.58 MB

#3

I actually do have separate images that stained for Hoechst:
Experiment-14-36 (dragged)-nucleus.tiff (1.0 MB) Experiment-10-24 (dragged)-nucleus.tiff (1.0 MB)
Not sure how much help they are though.

1 Like
#4

If the images were taken as CZI, it would generally be easier to work in the original format, but again, the rounded shape is likely based on how the cells are generated.
And you are right, the nuclear channel image is too fuzzy to get really good cell segmentation. I could at least tell where the cells were cytoplasmic positive for the red channel, but the counts would almost certainly be off due to segmentation errors.

image