Automating At-Home Balance Training Using Wearable Sensors

衰老和感觉障碍导致的平衡能力下降对生活质量和长期健康有负面影响。平衡不良会增加跌倒的风险、对跌倒的恐惧以及久坐不动的生活方式，这些都会导致随后的发病率、死亡率和医疗费用的增加。平衡训练旨在加强或恢复导致成功平衡的复杂感觉运动通路，可以改善有跌倒风险的个人的功能；然而，由理疗师进行的临床治疗受到患者负荷和保险限制的限制。目前的家庭培训并不像预期的那样有效，因为如果没有专家的指导，患者无法准确评估自己的表现或有效地自我提高培训。该项目将通过开发和验证可穿戴技术和数据驱动模型来推动科学发展，促进国家健康、繁荣和福利，这些可穿戴技术和数据驱动模型能够：1)远程评估用户家中的平衡；以及2)推荐由观察到的临床决策模型提供信息的平衡练习，这些模型根据用户不断发展的平衡能力充分和安全地挑战他们。这项研究是实现创建自动平衡训练技术以补充、补充和增加获得临床质量护理的机会这一长期目标的第一步，也是必要的一步。这项研究的结果有可能适用于具有包括感觉、神经和运动障碍在内的各种平衡和步态障碍的美国不同人群。此外，学生将通过多门课程参与，包括具有更广泛社区互动的动手设计课程项目。这项研究向开发可穿戴技术和机器智能迈出了第一步，这将使平衡训练计划能够在没有实时物理治疗师(PT)指导的情况下进行。该项目将(1)确定物理治疗师使用的重要运动学和视觉信息，以估计潜在的平衡运动能力，并为有关平衡运动进展的临床决策提供信息，以及(2)评估自适应机器学习模型的能力，以模拟专家知情的平衡运动进展战略，以响应不同个人和群体的不断变化的需求。为了实现这些目标，将以实时和异步视频记录的形式收集眼动跟踪和患者运动测量，以确定与物理治疗师对成人平衡能力评估相关的信息收集的关键方面。此外，强化学习框架下的马尔可夫决策过程建模将捕捉到物理治疗师-患者共同适应有效的运动进展策略的动态。待开发的模型将捕捉专家与患者互动的迭代、共同适应过程，该过程在培训计划的过程中发展，其中患者根据选定的培训调整他们的感觉运动行为，物理治疗师根据患者的进展调整培训。这项工作将导致开发整合不同临床和生物力学数据的模型，并生成新的方法来模拟专家与患者之间的交互，这些方法对患者差异具有健壮性，并随着时间的推移在专家和患者之间相互适应。在康复环境中描述物理治疗师与患者互动的动态过程的模型的开发，有望为未来为老年人和前庭功能障碍患者开发有效、可扩展的家庭平衡训练解决方案的努力提供信息。这些解决方案将补充和/或补充目前在临床环境中使用家庭自适应可穿戴技术提供的平衡训练。此外，通过这项研究开发的框架和技术可能适用于其他以临床为基础的培训环境(例如，中风恢复和手术后康复)。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Reductions in balance ability caused by aging and sensory disabilities have a negative impact on quality of life and long-term health. Poor balance increases the risk of falls, fear of falling, and sedentary lifestyles, which contribute to subsequent morbidity, mortality, and increased healthcare costs. Balance training designed to strengthen or restore the complex sensorimotor pathways that lead to successful balance can improve function in individuals at risk for falls; however, clinic-based sessions administered by physical therapists are limited by patient load and insurance constraints. Current home-based training is not as effective as it could be because patients cannot accurately assess their performance or effectively self-progress their training without expert guidance. This project will advance science and promote national health, prosperity, and welfare by developing and verifying wearable technology and data-driven models capable of 1) remotely assessing balance in users’ homes; and 2) recommending balance exercises informed by models of observed clinical decision-making that adequately and safely challenge users based on their evolving balance abilities. This research is a first and necessary step in achieving the long-term goal of creating automated balance training technologies to complement, supplement, and increase access to clinic-quality care. The outcomes of this research have the potential to be adapted to a diverse population of Americans with a wide range of balance and gait impairments including sensory, neurological, and motor disorders. Additionally, students will be engaged through multiple curricular offerings including hands-on design course projects with broader community interactions.This research takes the first step toward developing wearable technology and machine intelligence that will enable balance training programs that can be performed in the absence of real-time physical therapist (PT) guidance. The project will (1) identify important kinematic and visual information used by physical therapists to estimate underlying balance exercise ability and to inform clinical-decision making regarding balance exercise progression, and (2) assess the capabilities of adaptive machine learning models to simulate expert-informed balance exercise progression strategies that are responsive to the evolving needs of different individuals and groups. To achieve these objectives, eye movement tracking and patient kinematic measures will be collected in both live and asynchronous video-recorded formats to identify key aspects of information-gathering relevant to physical therapists’ evaluations of adult balance capabilities. Additionally, Markov decision process modeling under a reinforcement learning framework will capture the dynamics of the physical therapist-patient co-adaptation of effective exercise progression policies. The models to be developed will capture the iterative, co-adaptive process of expert-patient interaction that evolves over the course of the training program, where the patient adapts their sensorimotor behavior due to the selected training and the physical therapist adapts the training based on patient progress. This work will result in the development of models that integrate heterogeneous clinical and biomechanical data and generate new approaches for modeling expert-patient interaction that are robust to patient differences and co-adaptive between the expert and patient over time. The development of models that characterize the dynamic process of physical therapist-patient interaction in a rehabilitative setting promises to inform future efforts to develop effective, scalable at-home balance training solutions for older adults and people with vestibular dysfunction. Such solutions would complement and/or supplement the current provision of balance training within clinical settings using adaptive wearable technology at home. Furthermore, the framework and technology that will be developed through this research may be adapted for use in other clinic-based training contexts (e.g., stroke recovery and post-surgical rehabilitation).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.