PREPRINT: Introducing the Open and Reproducible Musculoskeletal Imaging Research (ORMIR) community

1 Preprint on Zenodo www.doi.org/10.5281/zenodo.8119062. 6 July 2023 Introducing the Open and Reproducible Musculoskeletal Imaging Research (ORMIR) community Serena Bonaretti 1 , Leonardo Barzaghi 2 , Mojtaba Barzegari 3 , Andrew J. Burghardt 4 , Donnie Cameron 5 , Julio Carballido-Gamio 6 , Gianluigi Crimi 7 , Pholpat Durongbhan 8 , Michelle Espinosa Hernandez 9 , Giulia Fraterrigo7, 10 , Lorenzo Grassi 11 , Hastings Greer 12 , Gianluca Iori 13 , Michael Kuczynski 14 , Sarah Manske14, Matthew McCormick 15 , Nathan Neeteson14, Marc Niethammer12, Martino Pani 16 , Jilmen Quintiens 17 , Majid Mohammad Sadeghi 18 , Francesco Santini 19 , Enrico Schileo7, Kathryn S. Stok8, Fulvia Taddei7, Justin J. Tse14, Jared Vicory15, Mariska Wesseling 20 , Andy Kin On Wong 21 , Dženan Zukić15 1 Swiss Center for Musculoskeletal Imaging, Balgrist Campus, Zurich, Switzerland 2 Advanced Imaging and Radiomics center, Neuroradiology Department, IRCCS Mondino Foundation, Pavia, Italy 3 Biomechanics section, Department of Mechanical Engineering, KU Leuven, Leuven, Belgium 4 Department of Radiology and Biomedical Imaging, University of California, San Francisco, United States 5 C.J. Gorter MRI Center, Department of Radiology, Leiden University Medical Center, Leiden, Netherlands 6 Department of Radiology, University of Colorado Anschutz Medical Campus, Denver, United States 7 Bioengineering and Computing Laboratory, IRCCS Istituto Ortopedico Rizzoli, Bologna, Italy 8 Department of Biomedical Engineering, The University of Melbourne, Parkville, Australia 9 Rehabilitation Sciences Institute, The University of Toronto, Toronto, Canada and Department of Biomedical Engineering, The University of Melbourne, Parkville, Australia 10 Department of Mechanical and Aerospace Engineering. Politecnico di Torino, Torino, Italy 11 Department of Biomedical Engineering, Lund University, Lund, Sweden 12 Department of Computer Science, The University of North Carolina at Chapel Hill, Chapel Hill, United States 13 Synchrotron-light for Experimental Science and Applications in the Middle East, Allan, Jordan 14 Department of Biomedical Engineering, McCaig Institute for Bone and Joint Health, Cumming School of Medicine, University of Calgary, Calgary, Canada 15 Medical Computing Department, Kitware, Inc, United States 16 School of Mechanical and Design Engineering, University of Portsmouth, Portsmouth, UK 17 Department of Mechanical Engineering, Biomechanics Section, KU Leuven, Leuven, Belgium 18 Department of Orthopedic Surgery, Maastricht University, The Netherlands 19 Basel Muscle MRI, Department of Biomedical Engineering, University of Basel, Basel, Switzerland and Department of Radiology, University Hospital Basel, Basel, Switzerland 20 Department of Biomechanical Engineering, TU Delft, Delft, the Netherlands 21 Joint Department of Medical Imaging, University Health Network, Ontario, Canada 2 Preprint on Zenodo www.doi.org/10.5281/zenodo.8119062. 6 July 2023 ABSTRACT In musculoskeletal (MSK) imaging research, scientists extract quantitative information from medical images to investigate chronic and debilitating diseases such as arthritis, osteoporosis, and neuromuscular diseases. The computational tools used to perform analyses are usually highly fragmented and often involve proprietary software, causing resource waste in re-implementations and ultimately slowing the advancement of the research field as a whole. The use of different code for evaluating similar processes also undermines scientific comparison and rigor. The ORMIR community aims to create and disseminate open, reproducible, well-tested, and well-documented software to analyze musculoskeletal images. Any interested scientists are welcome to join and contribute to the community. ARTICLE In musculoskeletal (MSK) imaging research, scientists extract quantitative information from medical images of joints, bones, and muscles to investigate chronic and debilitating diseases such as arthritis, osteoporosis, and muscular dystrophies (1–9). Computational tools used to perform analyses consist of workflows with a common structure: 1) acquisition of medical images using, e.g., computed tomography or magnetic resonance, either at the organ level (mm resolution) or tissue level (µm resolution in the order of); 2) segmentation of images to extract the MSK organs and tissues of interest—primarily bone, muscle, and cartilage; and 3) computation of metrics to quantify organ or tissue morphology, composition, and mechanical response. Variations across computational workflows are determined by implementation choices, including algorithms, computational parameters, and evaluation metrics. Both within and across research laboratories, the underlying development of computational workflows is subject to several challenges. In research groups, locally developed code is usually fragmented, as it mainly consists of a combination of proprietary software and in-house algorithms. Proprietary software is often distributed with pre-set pipelines; thus, scientists can rarely verify parameters and implementations, creating difficulties in adapting the code to new images or anatomies. In addition, in-house software is often created as part of a specific project of limited duration. Consequently, code life is strictly linked to the employment of the code creators, restricting code reuse and expansion by newer lab members. It can also reduce the integrity of scientific results when studies cannot be directly compared due to dissimilar analysis approaches, limiting research reproducibility and falsification. In the broader MSK imaging research community, software is usually not open source and, even when it is, it is often difficult to reuse because of lack of documentation (10), or it depends on proprietary languages and environments, such as MATLAB or Mathematica. Consequently, laboratories can rarely reuse existing code, unless they allocate resources to re-implementing algorithms from publications, which often lack implementation details (11,12). Further, absence of openly shared code at a community level limits compatibility with evolving technologies, expansion of code functionalities (13), comparison with new algorithms, and creation of validated standard procedures that can be trusted by the scientific community as a whole (14). Collaborations across laboratories would solve these challenges and potentially accelerate scientific discoveries in musculoskeletal research (15,16). 3 Preprint on Zenodo www.doi.org/10.5281/zenodo.8119062. 6 July 2023 Several research communities in other disciplines have sought to tackle code fragmentation by creating open software distribution platforms. In genomic research, ‘Bioconductor’ is a software distribution that collects more than 2,000 packages written in the R programming language to analyze data ranging from single-cell sequencing to flow cytometry (17). In geoscience research, ‘Pangeo’ is a high-performance-computing environment with core packages for big data research, showcased by a rich Jupyter notebook gallery (18); the Pangeo community is followed by more than 5,000 researchers on Twitter and by more than 800 people on its blog medium.com/pangeo. In the brain imaging community, researchers coding in R can benefit from ‘Neuroconductor’, an open-source platform for rapid testing and dissemination of computational imaging software, currently hosting 86 packages (19). Following the example of successful communities in other research fields, we established the Open and Reproducible Musculoskeletal Imaging Research (ORMIR) community, currently including more than 30 scientists from international academic institutions and industry. The proposal of code sharing, openness, and reproducibility as a solution to code fragmentation started circulating within the MSK imaging research community during satellite events 1 of the Quantitative Musculoskeletal Imaging (QMSKI) Workshop in 2019. A few months later, a group of nine researchers 2 took formal initiative and successfully applied for funds to hold a Jupyter Community Workshop. Originally planned for 2020, it was eventually held in 2022 due to the global pandemic (workshop report at (21)). Since the community was founded, the membership has increased in number, and members have started collaborating on computational workflows in four specific fields that address the musculoskeletal burden of arthritis, osteoporosis, and muscular dystrophies: quantification of bone and joint morphology from HighResolution peripheral Quantitative Computed Tomography (HR-pQCT) images, image-based micro finite element modeling, standardization of data formats for magnetic resonance (MR) muscle imaging, and analysis of MR images of the knee (Table 1). Initial results were disseminated during concomitant events 3 at the QMSKI 2022 workshop, and packages are currently under further development or maintenance. The ORMIR community aims to continue creating and disseminating open, reproducible, welltested, and well-documented software for analyzing musculoskeletal images. Future initiatives will include creating templates to homogenize code documentation and a certification system to standardize code. In the longer term, the community aims to share images and experimental data to create large datasets for cross-validation of algorithms and larger-scale computations. We will also seek to engage further with clinicians and clinical trial specialists to use these approaches to answer clinical questions relating to arthritis, osteoporosis, and neuromuscular diseases. The ORMIR community regularly shares updates on releases, meetings, and events through a website (ormircommunity.github.io) and a Twitter account (@ORMIR_Community), and aims to convene 1 “Hands-on Transparent QMSKI research: Open data, reproducible workflows, and interactive publications” organized by S. Bonaretti (20), and “Working group on standardization of quantitative metrics for 3D imaging” organized by A. Burghardt, P. Shneider, and S. Boyd 2 Barzegari M., Bonaretti S., Burghardt A., Carballido-Gamio J., Grassi L., Manske S., Schileo E., Stok K., Taddei, F. 3 “Introducing the Open and Reproducible Musculoskeletal Imaging Research (ORMIR) community” organized by S. Bonaretti, E. Schileo, and S. Manske (22), and “Are open and reproducible workflows necessary for CT in clinical trials and research” organized by S. Manske, A. Burghardt, and K. Stok 4 Preprint on Zenodo www.doi.org/10.5281/zenodo.8119062. 6 July 2023 in-person around the time of the QMSKI workshop every two years. We invite any interested scientists to join our ORMIR community and contribute to accelerating and supporting advancement of quantitative MSK imaging research. Python package name Description GitHub repository Packages developed with the support of the ORMIR community Ciclope Processing of micro computed tomography images to generate micro finite element models. github.com/gianthk/ciclope MuscleBIDS Reading and writing a standardized data format for muscle MR imaging that is based on BIDS github.com/muscle-bids/muscle-bids ORMIR_XCT Computing bone microarchitecture and joint space from HR-pQCT images github.com/SpectraCollab/ORMIR_XCT Pre-existing packages created and maintained by members of the ORMIR community pyKNEEr Segmenting and analyzing femoral knee cartilage from MR images github.com/sbonaretti/pyKNEEr OAI analysis 2 Segmenting, registering, and analyzing femoral and tibial knee cartilage from the MR images of the whole OAI dataset github.com/uncbiag/OAI_analysis_2 ITKIOScanco An ITK module to read and write Scanco microCT .isq files github.com/KitwareMedical/ITKIOScanco Dafne A segmentation tool for muscle MR images based on federated deep learning github.com/dafne-imaging Table 1. Python packages currently supported by the ORMIR community. Each GitHub repository contains the Python package and information on how to use it. Standardization of code style, documentation style, and file formats is under development. Abbreviations: MR: Magnetic Resonance; BIDS: Brain Imaging Data Structure (23); OAI: OsteoArthritis Initiative (24); HR-pQCT: High-Resolution peripheral Quantitative Computed Tomography. 5 Preprint on Zenodo www.doi.org/10.5281/zenodo.8119062. 6 July 2023 References 1. Díaz-Manera J, Llauger J, Gallardo E, Illa I. Muscle MRI in muscular dystrophies. Acta Myol Myopathies Cardiomyopathies Off J Mediterr Soc Myol. 2015 Dec;34(2–3):95–108. 2. Oei L, Koromani F, Rivadeneira F, Zillikens MC, Oei EHG. Quantitative imaging methods in osteoporosis. Quant Imaging Med Surg. 2016 Dec;6(6):680–98. 3. Löffler MT, Sollmann N, Mei K, Valentinitsch A, Noël PB, Kirschke JS, et al. X-ray-based quantitative osteoporosis imaging at the spine. Osteoporos Int. 2020 Feb;31(2):233–50. 4. Stok KS, Burghardt AJ, Boutroy S, Peters MPH, Manske SL, Stadelmann V, et al. Consensus approach for 3D joint space width of metacarpophalangeal joints of rheumatoid arthritis patients using highresolution peripheral quantitative computed tomography. Quant Imaging Med Surg. 2020 Feb;10(2):31425–31325. 5. Link TM, Kazakia G. Update on Imaging-Based Measurement of Bone Mineral Density and Quality. Curr Rheumatol Rep. 2020 May;22(5):13. 6. Pimentel A, Bover J, Elder G, Cohen-Solal M, Ureña-Torres PA. The Use of Imaging Techniques in Chronic Kidney Disease-Mineral and Bone Disorders (CKD-MBD)—A Systematic Review. Diagnostics. 2021 Apr 26;11(5):772. 7. Diaz-Manera J, Pichiecchio A, Santini F, Filosto M. Editorial: Imaging of neuromuscular diseases. Front Neurol. 2021 Dec 2;12:814579. 8. Carballido-Gamio J. Imaging techniques to study diabetic bone disease. Curr Opin Endocrinol Diabetes Obes. 2022 Aug 1;29(4):350–60. 9. Durongbhan P, MacKay JW, Schadow JE, Davey CE, Stok KS. Quantitative morphometric analysis in tibiofemoral joint osteoarthritis imaging: A literature review. Osteoarthr Imaging. 2023 Mar 1;3(1):100088. 10. Bonaretti S, Gold GE, Beaupre GS. pyKNEEr: An image analysis workflow for open and reproducible research on femoral knee cartilage. Olier I, editor. PLOS ONE. 2020 Jan 24;15(1):e0226501. 11. Vandewalle P. Code sharing is associated with research impact in image processing. Comput Sci Eng. 2012 Jul;14(4):42–7. 12. Freire J, Chirigati F. Provenance and the different flavors of computational reproducibility. IEEE Data Eng Bull. 2018;41(1):15. 13. Rule A, Birmingham A, Zuniga C, Altintas I, Huang SC, Knight R, et al. Ten simple rules for writing and sharing computational analyses in Jupyter Notebooks. Lewitter F, editor. PLOS Comput Biol. 2019 Jul 25;15(7):e1007007. 14. Kluess D, Soodmand E, Lorenz A, Pahr D, Schwarze M, Cichon R, et al. A round-robin finite element analysis of human femur mechanics between seven participating laboratories with experimental validation. Comput Methods Biomech Biomed Engin. 2019 Sep 10;22(12):1020–31. 15. Donoho DL, Maleki A, Rahman IU, Shahram M, Stodden V. Reproducible research in computational harmonic analysis. Comput Sci Eng. 2009 Jan;11(1):8–18. 16. Munafò MR, Nosek BA, Bishop DVM, Button KS, Chambers CD, Percie du Sert N, et al. A manifesto for reproducible science. Nat Hum Behav. 2017 Jan 10;1(1):0021. 17. Kern L. Bioconductor annual report for 2021. 2022; 18. Pangeo [Internet]. [cited 2023 Jan 5]. Available from: https://pangeo.io/ 19. Neuroconductor [Internet]. [cited 2023 Jan 5]. Available from: https://neuroconductor.org/ 20. Bonaretti S. Hands-on transparent QMSKI: Open access data, reproducible workflows, and interactive publications [Internet]. Available from: https://doi.org/10.5281/zenodo.2577616 6 Preprint on Zenodo www.doi.org/10.5281/zenodo.8119062. 6 July 2023 21. Bonaretti S, Cameron D, Kuczynski M, Iori G, on behalf of the participants of the workshop. Report on the Jupyter Community Workshop “Building the Jupyter Community in Musculoskeletal Imaging Research” [Internet]. Medium. 2022. Available from: https://blog.jupyter.org/report-on-the-jupytercommunity-workshop-77016ab1d49b 22. Bonaretti S. Introducing the Open and Reproducible Musculoskeletal Imaging Research (ORMIR) Community [Internet]. Available from: www.doi.org/10.5281/zenodo.6973540 23. Gorgolewski KJ, Auer T, Calhoun VD, Craddock RC, Das S, Duff EP, et al. The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments. Sci Data. 2016 Jun 21;3(1):160044. 24. Peterfy CG, Schneider E, Nevitt M. The osteoarthritis initiative: report on the design rationale for the magnetic resonance imaging protocol for the knee. Osteoarthritis Cartilage. 2008 Dec;16(12):1433– 41.