Computational Speech Analysis in Alzheimer's Disease and Other Neurocognitive Disorders

摘要： 神经认知障碍(NCDs)的早期和准确诊断对于计划、治疗和 研究转介，但需要时间和专业知识往往是初级保健提供者无法获得的。演讲和 在几种非传染性疾病的病程早期，语言经常受损。之前的研究已经证明 计算机语音分析(CSA)的诊断潜力及其在健康对照组和对照组之间的差异 轻度认知障碍(MCI)和阿尔茨海默病等疾病。然而，有几个 使CSA成为诊断可行的筛查工具必须采取的其他步骤。这项建议包括 职业发展计划，为申请人提供以下方面的培训、指导和经验 将CSA技术应用于临床的领域：1)计算语言学和副语言学，2) 疾病的纵向标志，以及3)新的传播技术的设计。作为这次培训的一部分， 学术和专业技能，包括研究中的道德，也将得到扩大。独一无二的资质 已经挑选了指导和咨询小组，以确保拟议的培训和 研究。 这项拟议的研究是一项前瞻性、纵向、观察性、队列调查，包括两项不同的研究 组。第一组是经过高度挑选并具有良好特征的健康对照研究队列， 阿尔茨海默病和MCI受试者(A组)。在A组中，一种 将改进机器学习算法以区分阿尔茨海默病和MCI与健康对照 使用CsA。多元回归和基于体素的形态测量将用于更好地理解 A组CSA指标的组间差异也有统计学意义。然后该算法的临床应用将被 在不同非传染性疾病患者(B组)的临床队列中进行评估，以减少频谱偏差 可能存在于先前的研究中。作为两组的子目标，可能进一步改进算法 还将检查纵向CSA措施的结果。总体目标是开发直观、 可靠和可重复性的基于CSA的临床测量，将它们与已建立的神经精神病学相关联 和成像标志物，确定它们在临床人群中的有效性，并确定它们是如何改变的 时间到了。因此，这项研究将验证特定的语音特征作为有用的诊断标记 并解释为什么这些标记物在患者组之间有所不同，这两组患者都是主要的 在阿尔茨海默氏症等非传染性疾病筛查中设计新的、易于实施的工具的步骤 疾病。

Abstract: Early and accurate diagnosis of neurocognitive disorders (NCDs) is critical for planning, treatment, and research referral, but demands time and expertise often unavailable to primary care providers. Speech and language are often impaired early in the disease course of several NCDs. Previous research has demonstrated the diagnostic potential of computer speech analysis (CSA), with differences between healthy controls and disorders such as mild cognitive impairment (MCI) and Alzheimer's disease. However, there are several additional steps that must be taken to make CSA a diagnostically viable screening tool. This proposal includes a career development plan providing the applicant with training, mentorship, and experience in the following areas in order to bring CSA techniques into clinical practice: 1) computational linguistics and paralinguistics, 2) longitudinal markers of disease, and 3) design of novel technology for dissemination. As part of this training, academic and professional skills, including ethics in research, will also be expanded. Uniquely qualified mentorship and advisory teams have been selected to ensure the success of the proposed training and research. The proposed study is a prospective, longitudinal, observational, cohort investigation of two distinct research groups. The first group is a highly selected and well-characterized research cohort of healthy control, Alzheimer's disease, and MCI subjects (Group A). In Group A, the performance and reproducibility of a machine learning algorithm will be improved to distinguish Alzheimer's disease and MCI from healthy controls using CSA. Multiple regression and voxel-based morphometry will be used to better understand what may drive group differences in CSA measures in Group A as well. Clinical applications of this algorithm will then be assessed in a clinic-based cohort of patients with different NCDs (Group B) in order reduce spectrum bias likely present in prior studies. As sub-aims in both groups, possible further improvement of the algorithmic outcomes with longitudinal CSA measures will also be examined. The overall objective is to develop intuitive, reliable and reproducible CSA-based clinical measures by correlating them with established neuropsychiatric and imaging markers, determining their efficacy in clinical populations, and determining how they change over time. As a result, this research will validate specific speech traits as useful diagnostic markers of neurocognitive disease and explain why those markers differ between patient groups, both of which are major steps towards the design of novel and easily implemented tools in the screening of NCDs such as Alzheimer's disease.