A Multimodal Deep Learning Based Approach for the Early Detection of Neurodegenerative Diseases

基于多模态深度学习的神经退行性疾病早期检测方法

• 批准号: 2102329
• 金额: --
• 依托单位: University of Bristol
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2102329
• 关键词: Multimodal; Deep; Learning; Based; Approach

Project description:Across industrialised nations, the population is aging. By 2050 individuals over the age of 65 will comprise up to 30% of the population of countries such as Italy and Germany. Increased age is associated with an increase in chronic diseases, including neurodegenerative diseases (ND). In the USA there will be an estimated 68% increase in the incidence of Parkinson's disease between 2010-2030 (630,000 increasing to 1.06 million) [1], and 40% for Alzheimer's (increasing from 5 million to 7.1 million) [2]. In-home healthcare platforms (e.g. SPHERE) could provide a method of monitoring the health of users, particularly for such diseases which typically have effects on human motion (e.g. gait, eye movement [5]). Such monitoring devices (e.g. RGB-D cameras) could potentially detect early signs of such ND, or help in monitoring their progression - perhaps aiding healthcare services in giving more timely and targeted medical assistance. The project aims to address three main questions:"Can generative adversarial networks (GANs) be used to effectively generate realistic, varied data that can subsequently be employed to categorise individuals as healthy or unhealthy with regards to neurodegeneration?" "What types of sensors (e.g. cameras, gyroscopes, inertial measurement units (IMUs), pressure mats) and data (e.g. gait and body motion, eye movements) are most informative and complementary for the consistent early detection of ND, and how can this data be most effectively gathered with respect to experiments?""What experiments should be designed to simulate early signs of ND, and how should different machine learning methods be designed and validated when trained with the data from this experiments?"Project aim: Develop a multimodal deep learning architecture capable of analysing the healthiness of a human being using data acquired from their gait, eye movement, and cognitive impairment.o Collect multi-modal data to establish what features are the best to use in this architecture to provide an early diagnosis of a specific ND

项目描述：在工业化国家，人口正在老龄化。到2050年，65岁以上的人将占意大利和德国等国人口的30%。年龄的增加与慢性病的增加有关，包括神经退行性疾病(ND)。在美国，2010-2030年间帕金森氏症的发病率估计将增加68%(63万增加到106万)[1]，阿尔茨海默氏症的发病率增加40%(从500万增加到710万)[2]。家庭保健平台(例如SPHERE)可以提供一种监测用户健康的方法，特别是对于通常会影响人体运动(例如步态、眼球运动)的这类疾病。这类监测设备(例如RGB-D摄像机)可能会检测到这种ND的早期迹象，或有助于监测其进展--也许有助于医疗服务部门提供更及时和更有针对性的医疗援助。该项目旨在解决三个主要问题：“生成性对抗网络(GAN)能否被用来有效地产生真实、多样的数据，这些数据随后可以用来将神经退行性变方面的个体归类为健康或不健康？”什么类型的传感器(如相机、陀螺仪、惯性测量单元(IMU)、压力垫)和数据(如步态和身体运动、眼球运动)对于ND的早期检测最具信息量和互补性，如何才能在实验中最有效地收集这些数据？“应该设计什么实验来模拟ND的早期迹象，当用这些实验的数据训练时，应该如何设计和验证不同的机器学习方法？”项目目标：开发一个多模式深度学习架构，能够使用从步态、眼睛运动、O收集多模式数据以确定在此架构中使用哪些功能最好，以提供特定ND的早期诊断