This is an open source platform for end-to-end artificial intelligence (AI) in computational pathology. It will enable you to use AI for prediction of any "label" directly from digitized pathology slides. Common use cases which can be reproduced by this pipeline are:
- prediction of microsatellite instability in colorectal cancer (Kather et al., Nat Med 2019)
- prediction of mutations in lung cancer (Coudray et al., Nat Med 2018)
- prediction of subtypes of renal cell carcinoma (Lu et al., Nat Biomed Eng 2021)
- other possible use cases are summarized by Echle et al., Br J Cancer 2021: https://www.nature.com/articles/s41416-020-01122-x
This pipeline is modular, which means that new methods for pre-/postprocessing or new AI methods can be easily integrated. For an extensive protocol including many example scripts, please see https://www.biorxiv.org/content/10.1101/2021.12.19.473344v1
Deepmed has been tested on both Windows Server 2019 and Ubuntu 20.04. It requires a CUDA-enabled NVIDIA GPU and a Python installation of at least version 3.8. In most cases, deepmed can then be installed by typing:
pip install git+https://github.com/KatherLab/deepmedIn some cases it may be necessary to install pytorch manually in order for it to recognize the system's GPU. To do so, please refer to the pytorch installation guide.
To build the project's documentation, we need to install a few more dependencies:
pip install sphinx sphinx_rtd_themeAfter that, we can build the documentation by invoking the Makefile or
make.bat in the docs dictory, i.e.:
make -C path/to/deepmed/docs htmlon Linux systems or
path\to\deepmed\docs\make.bat htmlon Windows. Afterwards, the documentation can be found in
docs/build/html/index.html.
Deepmed comes with a set of integration tests. These can be invoked by running
cd path/to/deepmed && python -m unittest