AddBiomechanics: Automatic Processing and Sharing of Human Movement Data

点击翻译按钮获取中文摘要

Title: “AddBiomechanics: Automatic processing and sharing of human movement data” Abstract: Movement related injuries and disorders affect most people at some point in their lifespan. Treatments are difficult to develop because we have limited ability to predict how a proposed treatment will change the neuromusculoskeletal dynamics of a patient. Machine learning approaches to predict patient responses to hypothetical treatments would radically shorten the development time for novel treatments, but we lack sufficient clean public data to apply these methods. Biomechanics data is too heterogeneous, decentralized, and small to be useful for modern machine learning techniques. This proposal describes a novel collaborative approach to create a large biomechanics dataset. The main bottlenecks preventing the sharing of biomechanics data are the very large time cost to manually process the data, and the lack of incentives for sharing. We propose to address both of these problems with a cloud-based data processing automation tool, which researchers can use for free if they agree to share the resulting de-identified data. To demonstrate the social viability of our approach, we have developed a prototype to partially automate the processing of biomechanics data, saving researchers some of the time they spend collecting and processing data. We have hosted this tool as a cloud application called AddBiomechanics, available for free if users are willing to share anonymized versions of any data they upload. Despite minimal advertising and including only an initial set of features, since our launch in early July 2022 researchers from over 130 universities already use the tool to process and share data, and have already collectively shared 10,000 motion capture trials totaling more than 80GB of data now in a unified, ML-ready format. The first aim of our proposal is to develop methods to automate more of the processing of biomechanics data, saving researchers up to 90% of the time they spend collecting and processing data. To further encourage sharing of data, our second aim is to improve our cloud-based tools, provide more support to our users, and advertise the tools more broadly within the community. This project has broad support in the biomechanics community, evidenced by letters of support from researchers at 8 institutions in 5 countries, and the resulting dataset will lay the foundation for machine learning breakthroughs in the analysis of human movement and prediction of treatment outcomes by reducing friction to share and aggregate movement data. This will increase innovation and improve the treatment of movement related injuries and disorders, enhancing quality of life for millions of people.