Qupath to assist Deep Learning approach

[khaled benaggoune]

Dear all,

I am new in image processing and my project is counting the percentage of tumoral cells with deep Learning approach, my question is : what is the best way to use qupath in this context ?

my naive steps are the Following :

1- read ndpi image

2- extract régions of interest of 2000 x 2000 pixels as PNG file

3- the expert annotate each file with the counting tool and save the points

4- use python script to generate masks from the text file of the annotations points

Now i want to split the 2000 x 2000 image in patches of 256 x 256 png format, how qupath do that ?

P.S : i tried qupath to export annotations and masks from points of counting tool but the script didn't work, i think it works just with annotations.

Thank you

[Pete]

There's not really any need to extract regions of interest at 2000 x 2000 pixels if the pathologist can have access to the .ndpi file.  You can create fixed sized rectangles using this script:

https://petebankhead.github.io/qupath/scripting/2018/03/09/script-create-fixed-size-region.html

Select it, right-click and set each rectangle to be locked... or do that in a script, something like (untested):

getAnnotationObjects().each {it.setLocked(true)}

Then you can ask the pathologist to annotate within that region.

The existing annotation export scripts on my blog don't include point annotations, but you can already save these as .csv from the GUI when using the counting tool.  I'm guessing you're using Python next; you might want to take more of the work outside of QuPath and do it in Python (e.g. convert the coordinates into whatever format you need for training).