Using QuPath to calculate cells with different stain in fluorescence


Hi everyone,

I’m learning QuPath and I cannot figure out the functions to count the cells based on the stain (Red Green and Yellow). The image type is Fluorescence (if that helps).
So far I manage to use Positive Cell Detection and by changing the intensity threshold I can select only the cells that are stained. But I am unable to proceed further in selectively separating the cells by stain color.

The ultimate goal is to automate QuPath to scan other slides to count the cells by stain color.

Thanks for your help.


If you are trying to classify cells that already exist, this is probably your best bet.
Cell detection in QuPath and most other software is generally based on nuclear detection - which is usually the blue channel. Without a nuclear marker that shows up in 100% of the cells it may be difficult to accurately segment cells. Even then it can be challenging in certain tissues.


I’ve tried using the protocol for multiplex analysis. However, I can’t get the program to recognize cells that are stain red.
I was able to adjust the threshold to detect green and yellow stained cells. However I cant get it to selectively detect just green or yellow cells.
Any suggestions?


With only a JPEG to go off of, it is hard to be sure on the forum (we do not have access to all of the information you do, only what you post) - but it looks like you have a two channel image. So you are picking up the green cells if you choose the green channel, and the red cells if you pick the red channel. Unless you have a third yellow channel, you are not picking up yellow cells, just green cells that also happen to have red in them.

Without a marker that is in 100% of the cells, you do not have anything for cell detection to work off of. It is a similar problem to trying to measure how much green is in a green cell using the green channel. You cannot do that appropriately since you can’t use a channel as it’s own area threshold (plenty of other posts on similar topics).

You may want to approach this a different way, measuring area based off of a pixel classifier, or using a simple thresholder and some scripting. At the moment though, it sounds like you are trying to use the cell detection inappropriately. If I misunderstood and you do have a channel that is present in 100% of your cells, then please do correct me.

Another option might be creating a channel (in FIJI for example, or via scripting) by adding the other two channels together… but that will likely run into intensity issues if your signal to noise ratio is low or your channels are too different in terms of intensity.

1 Like

Is there a way to share my data? I cant upload to this forum as the file size is too big.