SCH:INT: Collaborative Research: Semi-Automated Rehabilitation in the Home

随着美国人口老龄化，对中风和关节炎等衰弱疾病和伤害的有效和可获得的康复服务的需求日益增加。长期密集康复很难以一种可获得和负担得起的方式进行管理，因为它需要频繁前往诊所(通常由护理员支持)，并与康复专家进行重要的一对一时间。远程医疗和远程保健作为向更广泛的人口提供以家庭为基础的健康和健康的具有成本效益的方式而日益突出。然而，自动化远程康复目前还不可行，因为治疗师的专家职能还不能完全自动化并在家里复制。此外，伸缩技术辅助的家庭康复还面临着重大的技术、行为和临床挑战。该项目旨在通过开发家庭半自动康复系统(Sarah)来应对这些挑战。该系统被定义为半自动的，因为它依赖于治疗师的远程参与来开发和调整治疗计划。Sarah系统使用远程治疗师的指令来指导患者在家中进行日常强化治疗。该系统使用非侵入性的廉价传感技术，注意患者的隐私，记录和分析患者的日常治疗过程以及患者在家中的一般活动。然后，Sarah系统根据患者的治疗活动和在家中的一般活动向他们提供反馈。该系统还向远程治疗师提供患者进展的摘要，以便他们可以调整程序以适应后续的治疗会话。第一个版本的Sarah系统专注于家庭中的上肢中风康复，因为研究团队在这一领域拥有重要的经验。该项目的其他产出，包括开发通用系统和相关方法，旨在支持各种以家庭为基础的康复背景。该项目的技术目标是开发融合基于知识和数据驱动的方法的运动评估算法。这种融合的方法在基于家庭的治疗过程中产生自动的患者评估反馈，以及患者治疗和日常活动的摘要，以帮助治疗师做出远程决策。该项目利用一个接近治疗师决策过程的分层贝叶斯模型(HBM)作为开发综合网络-人类运动评估算法的公共框架。治疗过程使用两个摄像头和四个可穿戴的惯性测量单元(IMU)来捕捉，而日常活动只能通过IMU来跟踪，以估计佩戴者的3D运动学。该项目将临床医生关于治疗任务和片段的专业知识与视频和IMU数据融合在一起，以在家中实现治疗的自动分割和评级。融合的治疗数据的网络-人类评估用于将日常生活活动中的低级别IMU特征跟踪转换为日常运动摘要，以帮助远程治疗评估和定制。自动摘要包括：治疗依从性、治疗效果的质量、患者每天在家里的活动和活动的数量、受损肢体的使用、在日常活动中检测到的任务以及对识别的信心。基于知识和数据驱动的计算运动分析方法的融合，以及计算机-人类设计过程本身，将产生更高水平的概括性见解，扩展到数据受限情况下的机器学习和深度学习的更多应用。该项目生产的低成本传感器网络和可穿戴传感器解决方案将提供实用的方法来监控真实世界环境中的运动学，例如改进的假肢和外骨骼控制系统，通过生物反馈防止工作场所伤害，以及增强人与机器人的合作。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。

With the aging of the US population, there is an increasing need for effective and accessible rehabilitation services for debilitating illnesses and injuries such as stroke and arthritis. Intensive long-term rehabilitation is challenging to administer in an accessible and affordable way as it requires frequent trips to the clinic (usually supported by a caregiver), and significant one-on-one time with rehabilitation experts. Telemedicine and telehealth are gaining prominence as cost effective ways to deliver home-based health and wellness to wider populations. However, automated tele-rehabilitation is not currently feasible as the expert functions of the therapist cannot yet be fully automated and replicated in the home. In addition, there are significant technical, behavioral, and clinical challenges to scaling technology assisted home-based rehabilitation. This project aims to address these challenges through the development of a system for Semi-Automated Rehabilitation At Home (SARAH). The system is defined as semi-automated because it relies on the remote participation of the therapist for developing and adapting the therapy program. The SARAH system uses the remote therapists’ instructions to guide the patient through daily intensive therapy sessions at the home. Using inexpensive sensing technologies that are non-intrusive and mindful of the patient’s privacy, the system records and analyzes the daily therapy sessions as well as the general activities of the patient in the home. The SARAH system then provides feedback to the patient based on their therapy activities and general movements around the home. The system also provides summaries of patient progress to the remote therapist so that they can adapt the program for subsequent therapy sessions. The first version of the SARAH system focuses on upper extremity stroke rehabilitation at the home as the team of researchers has significant experience in this space. Additional outputs from this project, including the development of a generalized system and relevant methodology, are designed to support a wide variety of home-based rehabilitation contexts. The technical goals of the project are the development of movement assessment algorithms fusing knowledge based and data driven approaches. This fused approach produces automated patient assessment feedback during home-based therapy, and summaries of patient therapy and daily activities to assist the therapist with remote decision making. The project utilizes a Hierarchical Bayesian Model (HBM) approximating the therapist decision process as a common framework for the development of integrative cyber-human movement assessment algorithms. Therapy sessions are captured using two video cameras and four wearable Inertial Measurement Units (IMUs), while daily activity is only be tracked through the IMUs to estimate the wearer's 3D kinematics. The project fuses clinician’s expert knowledge of therapy tasks and segments with video and IMU data to implement automated segmentation and rating of therapy at the home. The fused cyber-human assessment of therapy data is used to inform the translation of low-level IMU feature tracking during daily life activities into daily movement summaries assisting remote therapy assessment and customization. The automated summaries include: therapy adherence, quality of therapy performance, quantity of patient daily activity and movement in the house, use of impaired limb, tasks detected during daily activity, and confidence of identification. The fusion of knowledge based and data driven approaches for computational movement analysis, as well as the cyber-human design process itself, will yield higher-level generalizable insights extending to many more applications of machine learning and deep learning in data-constrained scenarios. The low-cost sensor networks and wearable sensor solutions produced by the project will provide practical ways to monitor kinematics in real-world environments such as improved control systems for prosthetics and exoskeletons, prevention of workplace injuries through biofeedback, and enhancements in human-robot collaboration.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.