We collected together relevant software from EPFL and worked on making this easily accessible to analysis in clinics by developing the “Imaging Server Kit” which is since being developed further.
This plugin can be dragged and dropped into QuPath, a standard software in this field, and allows almost any imaging algorithm to run remotely on data loaded in QuPath, dramatically improving time for running and testing algorithms for detecting cells for example.
https://github.com/Imaging-Server-Kit/
We have developed tools to assess cancer proliferation, focusing primarily on Ki67-stained immunochemistry (IHC) images of breast cancers at low resolution (10x, 20x), viewed as sets of cells or cell graphs to remove the dependence on image patch size of traditional computer vision approaches. The Ki67 proliferation indicator results from 3 key components, namely separate identifications of Ki67+/-, cancer cells and Ki67 hotspots, for which we propose competitive solutions both in terms of accuracy and speed. See this preprint for an
example: https://arxiv.org/abs/2603.00143
In close collaboration with the CHUV IT department, we compiled best practice resources for secure and reproducible containerization aligned with the requirements of clinical infrastructure and Trusted Research Environments like CHUV’s Chorus. These recommendations informed the developments in WP1 and were published in open access. Furthermore, this collaboration enabled us to gain an overview of CHUV’s current IT infrastructure and governance, which allowed the establishment of strategies for algorithm deployment.
These recommendations are available here:
https://imaging-plaza.epfl.ch/faq
https://swissdatasciencecenter.github.io/best-practice-documentation/docs/category/security/
