MICCAI 2020

3 MONAI v0.3 19 DAILY Mo n d a y (NVidia), Eric Kerfoot (Kind’s Collect London), Wenqi Li (NVidia), Annika Reinke (German Cancer Research Center, Hiedelberg), Hans Johnson (University of Iowa), and Matt McCormick (Kitware) . Additionally, I have been developing open source software since 1999, when my group at The University of North Carolina participated in the creation of ITK, and I have continued to expand my commitment to open science ever since joining Kitware in 2005, including chairing the MONAI external advisory board. The goal of MONAI is rooted in open science: MONAI seeks to accelerate the pace of research and development by providing a common software foundation and a vibrant community that promote and support deep learning in medical imaging. As stated on the http://www.monai.io website: “MONAI is a freely available, community-supported, PyTorch- based framework for deep learning in healthcare imaging . It provides domain-optimized foundational capabilities for developing healthcare imaging training workflows in a native PyTorch paradigm.” The MONAI platform continues the open science tradition that was particularly well promoted by ITK. The ITK platform was one of the original platforms for open science in the medical image analysis field, and it enjoys one of the largest communities of users and developers in that field. With ITK v5.1, the full ITK platform is available with pythonic wrapping ( pip install itk ), and ITK is integrated into MONAI v0.3 to support image I/O and enable image preprocessing (registration, regularization, and enhancement) as well as data augmentation methods specific to medical images (augmentation by inter-patient registration). MONAI v0.3 was released in October 2020. See https://docs.monai.io/en/ latest/highlights.html. This release includes numerous advancements that enhance the capabilities and performance of the toolkit as well as its openness: ease of use, documentation, and organization. The outstanding performance of MONAI arises from it being focused on medical imaging. For example, it provides support for a wide range of medical imaging data formats and modalities, including support for reading DICOM objects and reconstructing DICOM volumes using ITK wrapped for Python. Additionally, it provides network architectures that are highly suitable for biomedical applications such as the popular UNet architecture and variants on it. Perhaps most importantly, MONAI also provides a multitude of data augmentation methods, loss