TOPIC 425: POSTURE ANALYSIS THROUGH MACHINE LEARNING (PATHML) PHASE II

低成本视频的提供具有巨大的潜力，可以提供关于身体行为如何与健康相关的新见解，定义临床试验结果，并评估家庭内的身体功能， 研究和临床环境。新冠肺炎大流行要求迅速转变，纳入基于视频的数据收集和临床访问，确立使用视频进行临床结果评估的可行性。然而，缺乏软件工具来自动化视频并将其整合到与健康相关的应用程序中。我们的第一阶段项目产生了一个概念验证的PathML软件原型 这包括几个足够精确的机器学习模型，以注释姿势、体力活动强度和环境特征。二期工程有两个目标。第一个目标是通过创新的机器学习方法来增强第一阶段PathML原型的功能，以注释与物理行为研究相关的分类和用户体验增强，这将导致 面向市场的工具，满足身体行为研究人员的需求。此外，该项目寻求扩展该软件以包括临床相关的COA的自动识别和评分(例如，30 第二次坐立测试、两分钟步行测试)，与在衰老、运动障碍、截肢者和癌症康复方面拥有专业知识的临床研究人员合作。

The availability of low-cost video has tremendous potential to provide new insight into how physical behavior is associated with health, define clinical trial outcomes, and assess physical function within the home, research and clinical environments. The COVID-19 pandemic required a rapid shift to incorporate video-based data collection and clinical visits, establishing the feasibility of using video for clinical outcome assessments (COA). However, there is a lack of software tools to automate and incorporate video into health-related applications. Our Phase I project resulted in a proof-of-concept PathML software prototype that included several sufficiently accurate machine learning models to annotate posture, physical activity intensity, and features of the environment. The Phase II project has two goals. The first goal is to enhance the functionality of the Phase I PathML prototype through innovative machine learning approaches to annotate relevant to physical behavior research taxonomies and UX enhancements that will result in a market-ready tool addressing the needs of physical behavior researchers. Additionally, this project seeks to extend the software to include automatic identification and scoring of clinically relevant COA’s (e.g., 30 second sit-stand test, two-minute walk test), partnering with clinical researchers with expertise in aging, movement disorders, amputees, and cancer rehabilitation.