Use of machine learning to quantify cognitive function in AD, FTD, and DLB

项目摘要/摘要 认知评估是诊断评估疑似患者的关键因素 神经退行性脑疾病的早期症状阶段，包括阿尔茨海默病(AD)， 额颞叶变性(FTLD)和路易体病(LBD)。随着生物标志物的成熟， 菲尔德正在将这些疾病引起的临床症状与神经病理变化分开 这导致了人们对这些疾病产生的疾病分类系统的担忧。 认知评估中通常使用的许多测试都是口头的，通常由 检查人员问病人一个问题，病人回答。这些测试通常按以下方式评分 手，考官计算正确或错误的答案，并产生一个简单的分数 与标准分数的对比。许多仍在使用的测试是在30年前开发的。一场激动人心的 人工智能方法的一系列最新进展已经开始使测量和 在录音中捕捉到的语言和其他认知特征的分类。 在这里，我们介绍一种实现这两种可能性的新方法。近期 自然语言处理(NLP)和机器学习(ML)的发展现已取得 有可能对语音样本中可测量的特征进行自动发现和分类。一次 建立后，这些特征集可以连接到皮质萎缩的分布，从而实现链接 在特定的认知异常和潜在的神经网络之间。这种分析方法 可以通过足够大量的认知测试来实现临床痴呆症综合征 作为语音样本记录的测量。此外，这种分析需要使用最新的 人工智能模型的生成，称为变压器网络，能够学习独特的 认知障碍或痴呆症患者回答以下问题的方式 记忆、执行功能、情感或语言。在目标1中，我们将调查一个 无监督人工智能模型用于测量AD患者的认知异常， FTLD，或DLB，反对传统的临床措施和区域性脑萎缩的MRI措施。 在目标2中，我们将研究一种无监督人工智能模型的性能 将AD、FTLD和DLB患者的认知异常归类为一组。在《目标3》中，我们将 评估这些自动测量AD、FTLD和DLB中认知异常的可靠性。 通过对AD、FTLD或DLB患者的认知进行更细粒度的分析，应该有可能 更好地了解这些痴呆症的重叠和分离特征， 以改进诊断分类和更好的监测。

Project Abstract / Summary Cognitive assessment is a key element of the diagnostic evaluation of patients suspected of having early symptomatic stages of neurodegenerative brain diseases, including Alzheimer’s disease (AD), Frontotemporal Lobar Degeneration (FTLD), and Lewy Body Disease (LBD). As biomarkers mature, the field is separating clinical syndromes arising from these diseases from the neuropathologic changes themselves, leading to concerns about classification systems for the illnesses these diseases produce. Many tests typically employed in cognitive assessment are verbal, often implemented by an examiner asking the patient a question and the patient answering. These tests are typically scored by hand, with the examiner counting correct or incorrect answers and a simple score being generated against normative scores. Many of the tests still in use were developed 30+ years ago. An exciting array of recent advances in artificial intelligence methods has begun to enable the measurement and classification of language and other cognitive characteristics captured in audio recordings. Here we introduce a new approach to accomplishing both of these possibilities. Recent developments in Natural Language Processing (NLP) and Machine Learning (ML) have now made possible the automated discovery and classification of features measurable in speech samples. Once established, these feature sets can be connected to distributions of cortical atrophy, thus enabling links between specific cognitive abnormalities and underlying neural networks. This approach to the analysis of clinical dementia syndromes can be achieved through a sufficiently large number of cognitive test measures recorded as speech samples. In addition, such analyses require the use of the latest generation of artificial intelligence models, called transformer-networks, to be able to learn the unique ways in which individuals with cognitive impairment or dementia respond to questions requiring memory, executive function, emotion, or language. In Aim 1, we will investigate the validity of an unsupervised artificial intelligence model for measuring cognitive abnormalities in patients with AD, FTLD, or DLB against traditional clinical measures and against MRI measures of regional brain atrophy. In Aim 2, we will investigate the performance of an unsupervised artificial intelligence model for classifying cognitive abnormalities in AD, FTLD, and DLB patients into clusters. In Aim 3, we will evaluate the reliability of these automated measures of cognitive abnormalities in AD, FTLD, and DLB. Through a finer-grained analysis of cognition in people with AD, FTLD, or DLB, it should be possible to develop better understanding of the overlapping and dissociable features of these dementias, aiming for improved diagnostic classification and better monitoring.