Whole slide glomerula detection

Hello everybody,

I am new to QuPath and played arround a little bit with this nice program (m9).
I managed to get my files from a zeiss z.1 into QuPath without problems.
With Analyze–> Preprocessing --> Simple Tissue Detection I get my tissue detected.
But now I struggle to go further.

I have 2 projects running:

  1. We have HE or PAS stained slides of kidney biopsies and I want to detect the number of glomerula inside my tissue and in this glomerula the number of stained nuclei.

  2. DAB stained kidney biopsies with different markers. We want to detect the stained nuclei inside the glomerula.

So my question is. Is there a way to autodetect glomerula in slide images? Some biopsies contain only 2 or 3. Others contain 40 glomerula.
Can s.b please point me to the right direction?

Best regards,

Hi @r100gs,

Have you tried cell detection? Here is the official documentation of QuPath, with the different types of detections/classifications that you can run!
My first guesses are (positive) cell detection (Analyze > Cell detection > (Positive) Cell detection), followed by object classifier (Classify > Object classification > Train object classifier) :slight_smile:
You could also try the pixel classifier if that corresponds better to your needs…
Good luck!

Detecting glomeruli is one of the poster child tricks for deep learning pixel classifiers (Visiopharm, Aivia), especially in HE staining. It might be fairly easy to use a pixel classifier with the other stains (especially PAS), but you will need a lot of texture features for HE. It is unlikely you will be able to use a single classifier for all three stains. Give the pixel classifier a shot first though, and you will likely need a decent sized training set with lots of features to get good results.

I think the last time I did this was using SLICs and classification, and the results were… ok but not great. If there are any deep learning pixel classifiers that can be run through ImageJ/FIJI and the resulting ROIs reimported into QuPath, that might be the most accurate way. Not sure how to actually do this though.

The most difficult one would be the DAB stained slides, as positive vs negative glomeruli are going to be difficult to handle. If your stains are sequential slices, you will probably want to align and transfer the glomerulus detections between images, based on whichever stain gives you the most accurate detection.

1 Like