VISTA integration in QuPath

Dear Community,
I am very fond of the QuPath interface, as it is quite intuitive and powerful. As the suite is extensible, I would be interested in how to incorporate the following additional tool in the suite: namely the VIsual Semantic Tissue Analysis (VISTA) tool for pancreatic diseases (VISTA: VIsual Semantic Tissue Analysis for pancreatic disease quantification in murine cohorts) would be a very useful segmentation algorythm for me. The code is freely available, but only for mac and linux GitHub - GelatinFrogs/MicePan-Segmentation.
Any ideas on how to proceed in a windows system? Thanks in advance!

Hi @pethozo,

I wasn’t aware of VISTA, but I’ve had a look at the code. I’m afraid it looks like it would be very difficult to link directly with QuPath at the moment. The code on GitHub is not really compatible: partly because it’s written in Python (QuPath is written in Java), but also because it doesn’t have an open source license (so the terms under which it can be used are unclear).

I’m also apprehensive about the fact only a small amount of code is available (for making predictions), rather than enough to actually recreate the AI models they use or to see precisely how it was done.

However, work we’re currently doing in QuPath that might help make methods such as this more easily available in the future. I’ve just been preparing my talk for the European Society of Integrative and Digital Pathology meeting this week and some of it is quite relevant.

3 Likes

Hi @petebankhead ,
Thank you for the feedback and for letting me know about the workshop yesterday! Your presentation was really great and indeed, I could very well see that the new features of QuPath (e.g. deep learning) could be useful for our future classifications.
The main take home message from yesterday’s workshop from my perspective was, definitely, to not only publish “representative” pictures from the slides but upload the whole scans to an online repository so that you and other could improve the classification algorithms. As I am new to histopathology and am overwhelmed by the sheer amount of online databases, could you recommend a repository that has a best availability from your side?

Hi @petebankhead

Is the content of your talk available somewhere?

Ah, thanks for joining @pethozo!

I used Google’s Dataset Search to find the public images I used in the presentation: https://datasetsearch.research.google.com

Although there seem to be a lot of public repositories of pathology slides (e.g. here), in practice I can rarely find a way to use them. They often only display images and seem to be focussed on teaching. For use in machine learning, we need to be able to download the images – and also for the slides to be available under a (usually Creative Commons) license that specifies how we can use them. The examples I got were hosted via Zenodo and Dryad.

I think it will be made available to ESDIP members, but I don’t think it will be publicly online. Although I have a suspicion that I will reuse at least some of the content in future talks, which could end up online.

(The most interesting stuff remains a preview, which we are too busy working on to be able to support yet – but which will come in a future release.)

2 Likes