Tissue detection for fluorescence images

Hi everyone,

I’m currently trying to detect tissue from fluorescence images. I understand that the ‘simple tissue detection’ function in QuPath detects tissues use the first channel so that the detected tissues are not always correct (even I set the threshold to very low, some tissues are still not detected). I note a script that provides more flexibility in channel selection to detect tissues, as shown in here: https://gist.github.com/Svidro/6171d6d24a85539d3af5d417bc928d50#file-tissue-detection-with-gui-groovy

Is this the right way to do simple detections for fluorescence images?

Thank you

If you are using 0.2.0 it is probably more straightforward to either train a pixel classifier, or if you want more control:
Create N number of Simple thresholders. Run them to create annotations and tag these two lines at the end.

selectAnnotations()
mergeSelectedAnnotations()

The above assumes there are no other annotations, since you generally start with tissue detection. I would probably opt for the second option as I don’t entirely trust pixel classifiers to be flexible enough as the various fluorescent channels vary across the tissue. If your nuclear density is high enough you might be able to get away with a decent Sigma blur and only the nuclear channel.

1 Like

I really like Morphological closing for this, actually, though you might get a more regular looking outline with Gaussian. Gaussian tends to cause problems for me if there are areas of low nuclear density mixed with areas of high nuclear density. Give them both a try.


1 Like

@HAOYANG_MI I recommend Classify → Pixel classification → Create thresholder.
It really replaces Simple tissue detection, and should be much more flexible – including for fluorescence.

See https://qupath.readthedocs.io/en/latest/docs/tutorials/thresholding.html

1 Like

@petebankhead @Research_Associate Thank you! This way works perfect for me, especially when combined with a minimum setting to the hole size and object size.

1 Like