Distinguish areals with carcinom and without

Hello, dear colleagues.
My name is Eugen, i’m pathologist and want correlate expression level of some IHC and density of CT
and/or PET by prostata cancer.
Could you please help me with:

  1. distinguishing aseals with prostatic adenocarcinom from normal tissue?
    Should i use “cell detection” and then “object classifier” with some different parameter? I already seen teaching video about using “nuclear/cytoplasm ratio”,but measurement of nuclear/cytoplasm ratio didn’t help me because for prostata CA exist another criteria: prominent nucleoli and absend of basal membrane…
    Is it normal that QuPath take up to 50-60 Gb aktiv memory at the time of cell detetion?
    My alternative variant was mark tumor and non tumor manually and then run pixel classifyer…but pixel classifyer work on the whole slide and analyse also areals, then i don’t want :wink: (nesty place).
    But generally it’s looks in the way they i want!

  2. Could i run pixel classifyer just on some ROI from annotation and not on the whole slide?
    Or like alternative way after pixel classification ignore areals without tumor - it will be also nice :slight_smile:

I add some pictures to visualize my problem :slight_smile:

Thank you in advance!

Hi Eugen, I’ll try to answer the technical bits… and avoid the parts that would require more pathology expertise :slight_smile:

Not for me, because I don’t have that much memory :slight_smile: But QuPath will use whatever resources it is allowed. I assume in your case you have a particularly powerful computer.

Some options to control this:

You can set the Region value when training the pixel classifier – this can be used to restrict the classification to only annotated areas. I would definitely recommend doing this when training a classifier at a high resolution, like in your case.

Note that it will still predict some pixels beyond the boundaries of annotations because the prediction is on a per-tile basis… and it will predict ‘all the tiles that overlap the bounding box of the annotation’.

However, if you make measurements / create objects, these can be restricted to only existing annotations. It should at least help exclude some irrelevant areas.

1 Like