A computational approach for quantifying motor behaviors in spinocerebellar ataxias to improve early detection of motor signs and precisely estimate disease severity and disease change

摘要 脊髓小脑性共济失调(SCA)是一种衰弱的神经退行性疾病，影响一系列 人类的行为包括手臂功能、言语和视觉。可以量化运动缺陷的工具 需要细粒度和客观的方式来支持对临床疾病发病的早期识别，更多 敏感地确定治疗的有效性，并对疾病的进展做出个性化的预测。 这些工具对于即将到来的SCA疾病修改临床试验是必要的，以减少样本量 和试验持续时间，并更好地了解给定的治疗如何改变人类行为。已关闭电源 目前这些罕见共济失调的主要预后指标，临床试验可能面临患者 招募和保留方面的挑战，特别是在多项临床试验同时进行的情况下。这些挑战可能 阻碍或减缓我们为患者成功发现治疗方法的能力。 我们最近在从语音中捕获多模式行为信号方面取得了实质性进展， 使用日常技术的眼球运动和手臂运动功能：麦克风、iPhone摄像头和 电脑鼠标。我们的初步数据表明，这些可伸缩技术具有很强的扩展潜力 目前对共济失调的临床评估以及我们用于产生疾病的新的机器学习方法 严重性估计比传统的回归模型方法执行得更好。我们的算法能够 在SCA患者的言语和手臂运动中定量识别共济失调和帕金森症的迹象，甚至 临床评估缺勤时。此外，我们的新的严重性估计算法使 对疾病进展的测量比临床量表更敏感。我们建议大幅扩大 纵向数据收集，并进一步开发我们的新分析方法，以训练更强大的模型 描述和量化人类的运动行为。所开发的技术有可能 促进临床试验，旨在为患有SCA的患者带来疾病修正疗法。虽然重点是 这个项目是关于SCA的，新的方法方法和产生的数据适用于其他 影响运动和语言的神经退行性疾病。此外，这个项目将带来新的洞察力 了解运动异常最初是如何产生和发展的。 该项目的总体目标是开发广泛可用的系统，以改进早期检测 脊髓小脑性共济失调的临床发病、严重程度评估和预后 同时了解这些障碍如何影响细粒度的运动行为。

ABSTRACT The spinocerebellar ataxias (SCA) are debilitating neurodegenerative diseases that impact a range of human behaviors including arm function, speech, and vision. Tools that can quantify motor deficits in a granular and objective manner are needed to support early recognition of clinical disease onset, more sensitively determine efficacy of a therapy, and make personalized predictions about disease progression. Such tools are needed for upcoming disease modifying clinical trials in SCAs, in order to reduce sample size and trial duration and better understand how a given therapy modifies human behaviors. Powered off of the currently available primary outcome measures for these rare ataxias, clinical trials are likely to face patient recruitment and retention challenges, especially with multiple co-occurring clinical trials. These challenges may impede or slow our ability to successfully discover therapies for our patients. We have recently made substantial progress in capturing multimodal behavioral signals from speech, eye movement, and arm motor function using everyday technologies: a microphone, iPhone camera, and computer mouse. Our initial data indicate that these scalable technologies have strong potential to extend current clinical assessments in ataxia and that our novel machine learning approach for generating disease severity estimates performs better than the traditional regression model approach. Our algorithms are able to quantitatively identify signs of ataxia and parkinsonism in SCA individuals' speech and arm movement, even when absent on clinical assessment. Furthermore, our novel severity estimation algorithm enabled measurement of disease progression more sensitively than clinical scales. We propose to substantially expand longitudinal data collection and further develop our novel analytic approaches to train more powerful models for characterizing and quantifying human motor behavior. The technologies developed have the potential to facilitate clinical trials aimed at bringing disease modifying therapies to individuals with SCA. While the focus of this project is on SCA, the novel methodological approaches and data generated are applicable to other neurodegenerative diseases affecting movement and speech. Furthermore, this project will bring new insight into how motor abnormalities initially arise and progress. The overall goal of this project is to develop widely available systems for improving early detection of clinical disease onset, severity assessment, and prognostication of spinocerebellar ataxias while simultaneously learning how these disorders impact fine-grained motor behavior.