Tissue detection using structure information

[kwiechen]

Hello,

I have tried tissue detection using the "Simple Tissue Detection (deprecated)" and "Create Thresholder" commands. The main problem here are often missing parts of adipose tissue. Is it possible to integrate another algorithm using structure information? Described for example in
Bug, Daniel, Friedrich Feuerhake, und Dorit Merhof. „Foreground Extraction for Histopathological Whole Slide Imaging“. In Bildverarbeitung für die Medizin 2015, 419–424. Springer, 2015. http://link.springer.com/chapter/10.1007/978-3-662-46224-9_72.

I have added some lines of code for testing to "SimpleTissueDetection2.java". The Laplacian operator was replaced with the Canny edge detector and the floodFill steps by findContours/fillPoly using the OpenCV bindings.
patch-tissue-extraction.zip

Best regards,

Kai

[petebankhead]

Hi Kai,

Thanks for the link, I only skimmed the first part of the paper. The method looks very interesting, I'll give some more thought about if/how it could be added to QuPath.

In the meantime, you can already integrate whatever algorithms you like via scripting (in Groovy) or creating a Java extension. Changing/adding core algorithms raises a lot more complications in terms of user expectations and reproducibility (QuPath has thousands of users...), not to mention maintenance, and increasingly we need to think carefully about algorithms that handle images with different numbers of channels.

Note that you can also use pixel classification, which provides more sophisticated options than simply thresholding.

I'd suggest reimplementing your changes as either a script or extension. Also, since this isn't a bug in QuPath, I will close the issue - the best place to discuss such things is the forum at http://forum.image.sc/tag/qupath

Best wishes,

Pete

[kwiechen]

Hi Pete, I have used the code from "Simple Tissue Detection" for demonstration purposes because I have no experience in Java (and Groovy too) development at all. The few lines of demo code are simply placed between your ui/image extraction code and the transformation of the tissue area mask to an annotation. Is it possible to reuse this module code and transfer to an extension or a script? Best regards, Kai Am So., 3. Jan. 2021 um 16:34 Uhr schrieb Pete <notifications@github.com>:

…

Hi Kai, Thanks for the link, I only skimmed the first part of the paper. The method looks very interesting, I'll give some more thought about if/how it could be added to QuPath. In the meantime, you can already integrate whatever algorithms you like via scripting (in Groovy) or creating a Java extension. Changing/adding core algorithms raises a lot more complications in terms of user expectations and reproducibility (QuPath has thousands of users...), not to mention maintenance, and increasingly we need to think carefully about algorithms that handle images with different numbers of channels. Note that you can also use pixel classification <https://qupath.readthedocs.io/en/latest/docs/tutorials/pixel_classification.html>, which provides more sophisticated options than simply thresholding. I'd suggest reimplementing your changes as either a script or extension. Also, since this isn't a bug in QuPath, I will close the issue - the best place to discuss such things is the forum at http://forum.image.sc/tag/qupath Best wishes, Pete — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#657 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEG6ZG4BEALLTARPZEDYZVLSYCFBTANCNFSM4VR5VYCA> .

[petebankhead]

Yes, for example see https://qupath.readthedocs.io/en/latest/docs/scripting/overview.html#working-with-opencv
There are many more scripting discussions on the forum.

[kwiechen]

I will give it a try - the QuPath script editor is a bit limited. Is there an alternative with syntax checking/intellisense .... Am So., 3. Jan. 2021 um 19:01 Uhr schrieb Pete <notifications@github.com>:

…

Yes, for example see https://qupath.readthedocs.io/en/latest/docs/scripting/overview.html#working-with-opencv There are many more scripting discussions on the forum. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#657 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEG6ZG53RJPZBSWJ2KTQ4HLSYCWGNANCNFSM4VR5VYCA> .

[petebankhead]

https://forum.image.sc/search?q=%22qupath%22%20%22intellij%22

[kwiechen]

Hello, I am trying to run the Tissue detection using structure information as a groovy script ending with a binary downsampled mask. I am struggling converting the mask information back to a QuPath annotation. Can you please give me a hint? Best regards, Kai Am So., 3. Jan. 2021 um 19:53 Uhr schrieb Pete <notifications@github.com>:

…

https://forum.image.sc/search?q=%22qupath%22%20%22intellij%22 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#657 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEG6ZGYKHKRM5ZB6XVXXQL3SYC4LVANCNFSM4VR5VYCA> .

[petebankhead]

As I mentioned in my first reply, the best place to discuss questions about QuPath is at http://forum.image.sc/tag/qupath for questions - I suggest posting your script there.

GitHub issues is only intended for bug tracking and we are trying not to have questions in multiple places.