Development of a Novel Integrative Recovery Index Using Machine Learning

在田径运动中取得成功通常需要精心设计的训练计划，在训练强度/体量和足够的休息/恢复之间取得平衡，以便进行渐进的身体和心理适应。制定训练计划的挑战是了解和评估训练负荷需要减弱的时间点。这降低了进入一种持久的、不受欢迎的疲劳状态的风险，这种状态被称为“功能过度”。目前还没有一个全面的指数来补充培训方案，以便在不影响恢复的情况下最大限度地增加培训负荷。然而，当代可穿戴智能手表技术显示出开发这样一个参数的前景。与目前可用的其他智能手表不同，由我们的行业合作伙伴Health Gauge开发的菲尼克斯智能手表能够捕获血压指标，并集成与情绪状态相关的心理定性反应选项。该项目的目标是与Health Gauge合作，将菲尼克斯智能手表的独特功能与机器学习(ML)相结合，开发出一种新颖、优化的训练恢复指数。该项目的第一阶段将包括将菲尼克斯手表校准为基于实验室的生物测量方法，包括在静态、动态和长时间(非卧床)条件下的大学运动员的心率、血压、心率变异性、温度和血氧饱和度。第二阶段将使用菲尼克斯从训练有素的自行车手那里获取生理和心理变量，以开发恢复指数，然后使用ML构建一个嵌入到手表中的框架，该框架将检测状态，并使用扩展的短期记忆模型将状态识别与训练输入相结合。一旦该模型完全开发出来，它预测“功能超负荷状态”的能力就可以在其他人群中得到验证。Health Gauge将把这一指数纳入他们的应用程序，使加拿大人能够访问它。除了监测运动员的准备情况外，我们还看到这一恢复指数有潜力指导紧急医务人员、医护人员、执法人员和军事人员的重返工作政策和建议。

Success in athletics usually requires well-designed training programs that balance training intensity/volume and adequate rest/recovery to allow for progressive physical and psychological adaptation. The challenge with developing training programs is to understand and assess the point at which training loads require attenuation. This reduces the risk of entering into a persistent and unwelcome fatigued state referred to as 'functional overreaching'. A comprehensive index that complements training programs to maximize training loads without compromising recovery currently does not exist. However, contemporary wearable smartwatch technology shows promise to develop such a parameter. Unlike other smartwatches currently available, the Phoenix smartwatch developed by our industrial partner Health Gauge is capable of capturing blood pressure metrics and integrating psychological qualitative response options related to mood states. The goal of this project is to work with Health Gauge to combine the unique capabilities of the Phoenix smartwatch with machine learning (ML) to develop a novel, optimized, training recovery index'. Phase 1 of the project will involve calibrating the Phoenix watch to lab-based biometric measures including heart rate, blood pressure, heart rate variability, temperature and blood oxygen saturation under resting, dynamic and prolonged (ambulatory) conditions in a sample of university athletes. Phase 2 will involve obtaining physiological and psychological variables from trained cyclists using the Phoenix to develop the recovery index with subsequent use of ML to build an embedding framework into the watch that will detect states and combine state recognition with training inputs using an extended short-term memory model. Once the model is fully developed, its ability to predict a 'functionally overreached state' can be validated in other populations. Health Gauge will incorporate this index into their application, making it accessible to Canadians. In addition to monitoring the readiness of athletes, we also see the potential for this recovery index to guide return-to-work policies and recommendations for emergency medical personnel, healthcare workers, law enforcement, and military personnel.