Democratization ofDeep Learning for Segmenting Cartilage from MRIs of Human Knees: Application toData from the Osteoarthritis Initiative.

In this study we aimedto democratize access to Convolutional Neural Networks (CNN) for segmenting cartilage volumes, generating state-of-the-art results for specialized, real-world applications in hospitals and research. Segmentation of cross-sectional and/or longitudinalMagnetic Resonance (MR) Images of articular cartilage facilitates both clinical management of joint damage/disease andfundamental research. Manual delineation of such images is a time-consuming task susceptible to high intra- and inter-operator variability and prone to errors. Thus, enabling reliable and efficient analyses of MRIs of cartilage requires automated segmentation of cartilage volumes. Two main limitations arise in the development of hospital- or population-specific Deep Learning (DL) models for image segmentation: specialized knowledge and specialized hardware. We present a relatively easy and accessible implementation of a DL model to automatically segment MR images of human knees with state-of-the-art accuracy. In representative examples we trained CNN models in six to eight hours and obtained results quantitatively comparable to the stateoftheart for every anatomical structure.We established and evaluated our methodsusing two publicly available MRI datasetsoriginating from the Osteoarthritis Initiative, Stryker Imorphics and Zuse Institute Berlin (ZIB), as representative test cases. We use Google Colabfor editing and adapting the Python codesand selecting the runtime environment leveraging high-performance GPUs.We designed our solution for novice users to apply to any dataset with relatively few adaptations requiring only basic programming skills. To facilitate adoption of our methods we provide a complete guideline for using our methods and software, as well as the software tools themselves. This article is protected by copyright. All rights reserved.