Using the QuPath pixel classifier to identify tissue types in Masson's Trichrome images

[Edward_Cui]

Hi there,

I’m trying to use QuPath’s pixel classifier to identify various tissue types (fibrous, smooth muscle, calcification, lipids etc.) in MT-stained histology slides of blood vessel plaque. My image type is set to “brightfield other.” To train the classifier, I have been manually drawing annotations for each tissue type/class on the slide, with the number of annotations per tissue type ranging from 3 to 25.

My current challenge is that I’m struggling to get the pixel classifier to distinguish between two tissue types (lipids and smooth muscle cells) that both appear as light purple/magenta regions, despite increasing the number of annotations and even loading training annotations from multiple slides. Since I know that lipids are more likely to be found in one part of the slide and less likely to be in other parts, is there a way to tell the pixel classifier where to look for a particular tissue type, or identify general regions that DO NOT contain a particular tissue type? Or is my best bet simply to increase the number of training annotations and hope the rate of misclassification decreases? Are there any image settings that would improve accuracy on Masson’s trichrome specifically?

I’m fairly new to QuPath, so any pointers and suggestions would be greatly appreciated!

[Research_Associate (MicroscopyRA)]

@Edward_Cui

is there a way to tell the pixel classifier where to look for a particular tissue type

Pretty sure no. Though as far as your second question, you can run classifiers within classifiers, so you could create annotations for the lipid+muscle areas, and then try and run a specific classifier to differentiate the two.

Hard to say too much without an example image for testing, but are you using texture based measurements from the list of filters?

From the readthedocs page. Pixel classification — QuPath 0.2.3 documentation

Also, SLICs are often more amenable to finding complex structures and the pixel classifier only has a certain level of “context” based on the 2/4/8x downsampling.

QuPath Guide: Creating classified annotation areas using SLICs

Quick tutorial on generating annotation level areas from SLICs. Image courtesy of La Jolla Institute of Immunology Microscopy Core shared image resource. Not going to go into a whole lot of details here as the entire method will likely be obsolete once the Pixel Classifier is fully functional. Generate your primary area with an annotation, either simple tissue detection, createSelectAllObject() or something else like drawing a square. Generate your SLICs. There isn’t any one right answer fo…

@Edward_Cui

Are there any image settings that would improve accuracy on Masson’s trichrome specifically?

As long as you are setting your color vectors correctly and using the color vectors instead of the regular RGB values, not much that I can think of.