Perturbation training for enhancing stability and limb support control for fall-risk reduction among stroke survivors

残留步态和平衡障碍是中风后跌倒和活动能力下降的主要危险因素。 因此，步态特征是慢性卒中患者的主要诊断和治疗目标 (PWCS)，准确和可重复的定量步态分析是该领域的关键。此外，平衡和步态 康复是使用最多的ICD 10(国际疾病分类)代码 神经功能障碍患者的康复。NICHD资助了几项针对新型药物的临床试验 平衡和步态干预，但该领域在数据共享、可获得性和 利用率。仪表化步态分析生成各种类型的大量相互依赖的数据；因此， 步态数据很难分析和解释。此外，人为的信息处理错误可能会导致 解释的可变性很高。这些障碍可以通过机器学习促进健康(ML4H)来解决 研究，其目标是创建模型，在有限或没有人工监督的情况下解决复杂任务。 不幸的是，最近的研究表明，ML4H在数据和代码方面不如其他ML字段 临床医生和研究人员的可访问性和公共步态数据库是有限的。即使 研究人员愿意共享数据，收集方法和文件结构的差异需要 任何人都可以使用这些数据的计算专业知识。我们可以访问从以下地点收集的大型数据集 R01HD088543：“增强稳定性的摄动训练和降低跌倒风险的肢体支撑控制 在中风幸存者中。“该项目是一项随机对照试验，测试PWCS获得、 推广和保留对滑移摄动训练的适应，不仅可以降低摔倒风险，还可以提高 步行功能。这项研究的假设如果得到结果的支持，将提供一个支持的证据 减少PWCS跌倒风险的训练方案。从授权生成的数据集包括运动学， 站姿控制、扰动和未扰动步态的动力学和临床措施(约1,500次试验 总计)。该合作项目旨在向数据驱动的民主化迈进一步 步态分析方法，为广泛的利益相关者提供支持(AL/ML研究人员，物理 治疗师、康复科学家)。在目标1中，我们将通过数据辩论和 协调能力(DataWrangler库)遵循公平数据原则(可查找性、可访问性、 互操作性和可重用性)。AIM 2将利用协调的数据集来创建科学的工作流 生物力学数据利用(GaitVis库，具有数据可视化、清理和分析功能)和 通过一个集中的网站(数据门户)提供对这些数据的访问。最后，在目标3中，我们将展示一个用例 用于开发基于步态数据和临床测量的预测跌倒风险模型。通过 通过共享术语使数据可被发现和机器可读，我们将支持未来 研究人员结合了不可预见的未来数据集，并提出了我们尚未考虑的问题。

Residual gait and balance impairments are major risk factors for post-stroke falls and reduced mobility. Therefore, gait characteristics are primary diagnosis and treatment targets for people living with chronic stroke (PwCS), and accurate and repeatable quantitative gait analysis is crucial to the field. Further, balance and gait rehabilitation are the most utilized ICD 10 (International Classification of Diseases) codes for physical rehabilitation of people living with neurological disorders. NICHD funds several clinical trials targeting novel balance and gait interventions, yet there is a gap in the field pertaining to data sharing, accessibility and utilization. Instrumented gait analysis generates a large amount of interdependent data of various kinds; hence, gait data are difficult to analyze and interpret. Furthermore, human information processing errors could lead to high variability in interpretations. Such barriers can be resolved via machine learning for health (ML4H) research, whose goal is to create models to solve complex tasks with limited or no human supervision. Unfortunately, recent studies show that ML4H compares poorly to other ML fields regarding data and code accessibility, and public gait data repositories for clinicians and researchers are limited. Even though researchers are willing to share data, differences in collection methods and file structures require computational expertise for anyone to use the data. We have access to a large data set collected from R01HD088543:“Perturbation training for enhancing stability and limb support control for fall-risk reduction among stroke survivors.” The project is a randomized controlled trial examining the ability of PwCS to acquire, generalize and retain adaptations to slip-perturbation training for not only mitigating fall risk but also improving walking function. The hypothesis of this study if supported by the results will provide an evidence-supported training protocol to reduce the fall-risk in PwCS. The data sets generated from the grant include kinematics, kinetics, and clinical measures for stance posture control, perturbed and unperturbed gait (about 1,500 trials total) from PwCS. This collaborative project aims to take a step towards democratizing data-driven approaches in gait analysis to empower a broad range of stakeholders (AL/ML researchers, physical therapists, rehab scientists). In Aim 1, we will evaluate and enable metadata through data wrangling and harmonization capabilities (DataWrangler library) following FAIR data principles (findability, accessibility, interoperability and reusability). Aim 2 will leverage harmonized data sets to create scientific workflows for biomechanical data utilization (GaitVis library, with data visualization, cleaning and analysis functionalities) and provide access this data through a centralized website (DataPortal). Lastly in Aim 3, we will show an use case of the transformed data for developing a predictive fall-risk model based on gait data and clinical measures. By enabling the data to be discoverable and machine readable through shared terminologies we will enable future researchers to combine unforeseen future data sets and ask questions we have not yet considered.