We use our own, openly available library faimed3d to build and train neural networks on medical data. faimed3d can be installed directly from GitHub with pip install git+github.com/kbressem/faimed3d. Alternatively one can also clone the repository and create a symbolic link to the faimed3d libs.
This project is implemented using nbdev, a true literate programming environment.
Magnetic resonance imaging (MRI) is used for early diagnosis of axial spondyloarthritis (axSpA). Diagnosis of axSpA requires a thorough knowledge of typical imaging findings and experience in rheumatologic imaging, which is challenging for non-specialized centers. Deep learning can facilitate and support diagnosis in clinical practice, but heterogeneity of MRI data and lack of reliable reference standards are important limitations of previous approaches. We have developed a deep learning tool that enables detection of changes characteristic of axSpA on MRI, overcoming the challenges associated with a large heterogeneous multicenter MRI dataset.
All code can be found in the nbs folder. The notebooks are sorted chronologically, according to the order in which the methodology is applied in the paper.
@article{bressem2022deep,
title={Deep learning detects changes indicative of axial spondyloarthritis at MRI of sacroiliac joints},
author={Bressem, Keno K and Adams, Lisa C and Proft, Fabian and Hermann, Kay Geert A and Diekhoff, Torsten and Spiller, Laura and Niehues, Stefan M and Makowski, Marcus R and Hamm, Bernd and Protopopov, Mikhail and others},
journal={Radiology},
volume={305},
number={3},
pages={655--665},
year={2022},
publisher={Radiological Society of North America}
}