v0.1.1 - model zoo, bug fixes

Bug fix release

In this release, a model zoo has been added, and numerous bugs have been fixed.

New functionality

• You can now make an additional_configs folder, and the config files in this folder will overwrite the configurations as given in the repository. This allows you to have a separate repository with your local configs without editing core (#7).
• You can now use Darwin V7 as a source for annotations (#33).

Bug fixes

• Update dlup to latest version (#20, #21, #44, #31) allowing (#29)
• Change compression for output TIFFs (#3)
• MPS acceleration support (#10)
• Support Darwin V7 (#34)
• Improve and fix database models (#14, #36)
• Fix segmentation models when ROI is absent (#27, #28, #39)
• Predict did not work before (#30, #37)
• Improve database (#14, #36)
• Improve H5FileImageWriter and TIFFWriter (#52, #49)
• Improve callbacks (#45)
• Fix CI/CD (#59)

Contributors

• @jonasteuwen
• @BPdeRooij
• @VanessaBotha
• @moerlemans
• @EricMarcus-ai
• @AjeyPaiK

Full Changelog: v0.1...v0.1.1

ahcore 0.1 - first version

First version of ahcore

Ahcore is the AI for Oncology public toolkit for computational pathology. It's goal is to eventually support all computational pathology workflows, such as segmentation, detection but also support advanced self-supervised pipelines and foundational models.

Features

Ahcore is a computational pathology toolkit, in the first public release we only support segmentation.

• Lightning AI-based computational pathology pipeline
• MONAI model support
• GPU-based augmentation pipeline based on Kornia
• Data loading supported by dlup
• Callbacks supporting the tile-by-tile inference and writing to TIFF
• Callbacks supporting to compute the whole-slide level metrics
• Hydra-based configuration pipeline.

Note: Ahcore v0.1 currently only supports segmentation, but detection will be added in the next version.

Documentation

A bit more documentation is available at https://docs.aiforoncology.nl/ahcore, and will be extended in the coming period, also including a few trained models.

Credits

Many members of the AI for Oncology lab were involved in preparing this version. Special thanks to (in no particular order):

• Eric Marcus @EricMarcus-ai
• Jonas Teuwen @jonasteuwen
• Vanessa Botha @VanessaBotha
• Marek Oerlemans @moerlemans
• Ajey Pai @AjeyPaiK

And the NKI's computational pathology lab:

• Hugo Horlings
• Bart de Rooij @BPdeRooij

Third-parties

Many thanks to all the authors and contributors of our dependencies, and the author of the wonderful lightning-hydra-template.