An automated approach for video-based motor assessment in Parkinson's disease

项目摘要 帕金森病（Parkinson's disease，PD）是一种神经退行性运动障碍，可导致一系列运动功能障碍。由于 PD的退行性本质，运动症状的有效管理需要频繁和准确的 运动评估目前用于PD运动评估的方法依赖于主观评定的临床量表或 昂贵或数据有限的研究级设备。这些限制极大地限制了 对运动功能进行频繁、精细的评估，并强调对新方法的重要需求 这使得在任何环境下都能以最少的时间、金钱或努力进行客观的运动评估。 在这里，我们提出了一个自动化的，基于视频的方法，PD运动评估，通过使用最先进的 位姿估计算法简而言之，姿态估计是一种新兴的技术，其能够定量地 从使用普通家用设备记录的简单数字视频跟踪人的运动（例如， 智能手机、平板电脑）。用于跟踪人体运动的姿态估计算法的发展 在人工智能社区中进展迅速，现在有巨大的未开发潜力， 利用该技术直接在家中或诊所进行快速、自动化的运动评估。 在目标1.1中，我们将测试用于PD患者步态评估的姿势估计方法。步态障碍是 在PD中很常见，但在家庭或诊所中定量评估步态仍然很困难。步态评估不是 在PD运动功能的临床量表中得到了充分的代表，并且没有很好的方法来客观评价 可直接用于家庭或诊所的全身步态运动学测量。我们将解决 这需要通过使用基于视频的姿势估计工作流程来执行PD患者的步态评估。 我们将这些结果与运动捕捉测量结果进行比较，以检查姿势估计 方法近似地面实况测量。 在目标1.2中，我们将使用我们的姿势估计方法来跟踪PD患者的重复运动。PD 经常导致难以进行重复运动（例如，手指敲击）;因此， 重复运动构成了MDS统一帕金森病评定量表的重要部分 (the用于PD中运动评估的标准临床量表）。这种评估是通过主观的人类 在这里，我们将测试一种自动化的，客观的，基于视频的方法来跟踪重复的动作， 患有PD的人，并再次与地面实况运动捕获测量进行比较。 在目标2中，我们将比较和对比姿势估计和人类评估识别的运动缺陷。 主观视觉评估是评估PD运动功能最常用的方法，但计算机 视觉方法（例如，姿态估计）具有使该过程自动化和客观化的巨大潜力。 在这里，我们将比较如何以及赤字步态和重复运动检测我们的姿态估计 这种方法与神经学家在统一帕金森病评定量表上对这些任务的评定一致。

Project Summary Parkinson’s disease (PD) is neurodegenerative movement disorder that causes a series of motor deficits. Due to the degenerative nature of PD, effective management of motor symptoms requires frequent and accurate motor assessment. Current approaches for PD motor assessment rely on subjectively rated clinical scales or research-grade equipment that is expensive or data-limited. These limitations significantly restrict the ability to perform frequent, granular assessments of motor function and highlight an important need for new approaches that enable objective motor assessments in any setting with minimal costs of time, money, or effort. Here, we propose an automated, video-based approach to PD motor assessment by using a state of the art pose estimation algorithm. Briefly, pose estimation is an emerging technology that is capable of quantitatively tracking human movement from simple digital videos recorded using common household devices (e.g., smartphones, tablets). The development of pose estimation algorithms for tracking human movement has progressed rapidly in the artificial intelligence community, and there is now significant untapped potential for leveraging this technology to perform rapid, automated motor assessments directly in home or clinic. In Aim 1.1, we will test a pose estimation approach for gait assessment in persons with PD. Gait dysfunction is common in PD, but it remains difficult to assess gait quantitatively in the home or clinic. Gait assessment is not represented adequately in clinical scales of motor function in PD, and there are no good methods for objective measurement of whole-body gait kinematics that can be used directly in the home or clinic. We will address this need by using a video-based pose estimation workflow to perform gait assessments in persons with PD. We will compare these results to motion capture measurements to examine how well the pose estimation approach approximates ground-truth measurements. In Aim 1.2, we will use our pose estimation approach to track repetitive movements in persons with PD. PD often causes difficulty in performing repetitive movements (e.g., finger tapping); accordingly, assessment of repetitive movements constitutes a significant portion of the MDS-Unified Parkinson’s Disease Rating Scale (the standard clinical scale for motor assessment in PD). This assessment is done through subjective human ratings; here, we will test an automated, objective, video-based approach for tracking repetitive movements in persons with PD and again compare to ground-truth motion capture measurements. In Aim 2, we will compare and contrast motor deficits identified by pose estimation and human assessment. Subjective visual assessment is the most common means of assessing motor function in PD, but computer vision approaches (e.g., pose estimation) have significant potential to automate and objectivize this process. Here, we will compare how well deficits in gait and repetitive movements detected by our pose estimation approach align with neurologist ratings of these tasks on the Unified Parkinson’s Disease Rating Scale.