Pixel classification from each fluorescence channel

I have fluorescence images that include three color channels, each of which labels one type cells. If I’d like to use “pixel classification” function to classify each type of cell, do I have to annotate cells from each channel in separate images and then train each classifier?

Hi @kylin,

I think you might be interested in the multiplex analysis section of QuPath docs. There’s an analysis pipeline there that will help you classify your objects according to different markers (channels), each with its own classifier. Good luck!


@kylin the multiplex analysis tutorial that @melvingelbard mentioned is the closest match to the application you describe, but note that it uses an object classifier and not a pixel classifier.

You don’t have to annotate cells in separate images – but it isn’t really possible to make useful suggestions without a lot more information about your images and what exactly you want to do with them.


The “duplicate channel images” mentioned in the tutorial is helpful but I ended up with annotating in separate images. For pixel classification, I created “Edge” to better classify the neighboring cell border and those "Edge"s are different for different fluorescence channels. Ultimately, it still works pretty well although it looks similar cell count compared to the “positive cell detection” which is quite subject to “intensity threshold” that has been set.