Pattern recognition for the detection and monitoring of verbal and non-verbal alterations in Alzheimer's disease

在加拿大，每十个65岁以上的成年人中就有一个患有阿尔茨海默病。在痴呆症的早期阶段进行干预可能会更有效。然而，特别是在低收入和中等收入国家，在疾病开始几年后诊断出阿尔茨海默病非常常见，导致早期痴呆症患者的治疗缺口。这种差距可能会降低治疗的有效性，延长患者独立性降低的状态。早期诊断和治疗现在被认为是降低护理成本和缩小这一差距的一种手段。此外，早期诊断将使患者和他们的家人能够更好地预测未来并做好相应的准备，从而使他们的事务井然有序。目前的计划旨在自动描述和监测AD的变化。这项研究侧重于分析言语和非言语行为的多种形式：言语、话语、面部和躯体表情。更具体地说，我们正在追求三个目标：1.使用从与老年受试者的对话的转录和音频来源提取的属性来区分AD患者和患病患者(言语行为分析)。使用从与老年受试者的对话录像中提取的属性来区分AD患者和患病患者(非言语行为分析)。构建一个机器学习系统，它将使用(1)和(2)中获得的每一个属性来识别AD的不同阶段。为了完成这些任务，我们使用了四个公认的数据集：西班牙BBVA基金会的BBVA、匹兹堡大学的DementiaBank Pitt语料库、南卡罗来纳医科大学的Carolina Conversations Collection和比利时的CorpAGEst语料库(欧盟第七框架计划)。前两个包含在对患者进行认知测试期间进行的对话。最后两个包含了在没有压力的环境中进行的自然的免费面试。涵盖的语言包括西班牙语、英语和法语。这项提议的影响是双重的。首先，它有助于建立更强大的面部和手势识别技术，以适应人口老龄化。世界各地的人口都在增长，这一贡献将使专门针对老年人的软件开发成为可能。其次，它有助于开发自动化工具来监控AD和其他类型的痴呆症(例如，药物对语言能力的影响)，为研究人员社区提供以自动方式探索特殊性的工具。这项提议的结果不是为了诊断或取代专家(例如临床医生、医生、老年医生)；它们的目的是以创新和非侵入性的方式帮助这些专业人员，帮助他们了解疾病及其进展。

Alzheimer's disease (AD) affects one in ten adults over 65 in Canada. Interventions may be more effective in the early stages of dementia. Nevertheless, it is highly common, especially in low- and middle-income countries, to diagnose AD several years after the disease begins, leading to a treatment gap for early dementia sufferers. This gap could reduce the effectiveness of treatments, prolonging patients' state of reduced independence. Early diagnosis and treatment are now recognized as a means of attenuating care costs and reducing this gap. Furthermore, an early diagnosis would allow sufferers and their families to get their affairs in order by better anticipating the future and preparing accordingly.The current program aims to automatically characterize and monitor changes in AD. This research focuses on analyzing verbal and non-verbal behaviours in multiple modalities: speech, discourse, and facial and corporal expressions. More specifically, we are pursuing three objectives:1. Distinguish between AD patients and ailing patients using properties extracted from transcriptions and audio sources of conversations with elderly subjects (analysis of verbal behaviours).2. Distinguish between AD patients and ailing patients using properties extracted from video recordings of conversations with elderly subjects (analysis of non-verbal behaviours).3. Construct a machine learning system that will use every property obtained in (1) and (2) to identify various stages of AD.To accomplish these tasks, we are using four recognized datasets: the BBVA of the BBVA Foundation in Spain, the DementiaBank Pitt Corpus from Pittsburgh University, the Carolina Conversations Collection from the Medical University of South Carolina, and the CorpAGEst corpus from Belgium (European Union Seventh Framework Program). The first two contain conversations made during the application of cognitive tests to patients. The last two contain natural free interviews in a non-stressful environment. The languages covered include Spanish, English and French.The impacts of this proposal are twofold. First, it contributes to building stronger facial and gesture recognition techniques adapted to the aging population. That population being on the rise everywhere in the world, this contribution will allow the development of software aimed specifically at the elderly. Second, it contributes to developing automated tools to monitor AD and other types of dementia (e.g., the effects of medications on language abilities), giving the community of researchers tools to explore specificities in an automatic fashion. The results of this proposal are not meant to diagnose or to replace specialists (e.g., clinicians, doctors, geriatricians); they are aimed at assisting these professionals in innovative and non-invasive ways, helping them to understand the disease and its progression.