SenSE: AI-Driven, Resilient and Adaptive Monitoring of Sleep (AI-DReAMS)

在美国和世界各地，有相当多的人患有慢性睡眠障碍。然而，其中大多数没有得到诊断或治疗。我们迫切需要新的可穿戴技术，以提高睡眠研究人员在研究睡眠、理解睡眠病理方面取得进一步进展的能力，并提高临床医生可靠地检测和治疗睡眠障碍的能力。该奖项旨在研究人工智能驱动的可重构睡眠监测系统的使用，以改变诊所和家庭的睡眠研究。由人工智能支持的传感器融合策略将使睡眠评估仪器超小型化，并探索新的睡眠相关生物标志物特征，这一策略具有革命性的潜力，可以激发睡眠研究，从而更有效、更准确地诊断和治疗睡眠障碍。将更低的成本与更好的人体工程学舒适性相结合，以及更高效的数据分析，将为这些技术在现实环境中的快速转换、采用和有效部署铺平道路。该奖项的研究成果有可能对其他慢性疾病（如心脏病）所需的连续监测仪器产生积极影响。该项目为推广和教育提供了几个激励机会，包括使用技术来连接睡眠，学习新的人工智能和数据分析技能，分析睡眠表现以解开其奥秘，以及睡眠对社会健康和效率的影响。计划利用这一研究项目的成果开展几项这样的活动，以便向年轻一代、教育工作者、外地的其他研究人员和广大公众提供服务。该奖项整合了硬件和数据分析创新方面的两项并行努力：1)在粘合剂贴片的形式因素中实现适应性和可重构嵌入式系统平台，2)开发最先进的机器学习技术，其中包含改善睡眠监测系统弹性所需的数据驱动模型。硬件系统融合了多模态可穿戴传感器，将近红外光谱（NIRS）与其他传统的睡眠相关信号传感器结合在一起，安装在符合皮肤的衬底上，以收集人体多个位置的数据。该系统的有效性将从提高一致性和灵活性、减少系统占用和成本等方面进行评估。数据分析平台包括1)信号处理，为给定推理任务的信号质量评估提供数据驱动指标；2)基于迁移学习技术和多种数据集的推理模型，用于使用新的传感模式检测睡眠事件和障碍；3)贝叶斯神经网络支持的传感器选择，用于提高睡眠传感器系统的弹性和适应性。“适应”将通过在设计迭代和操作过程中实时选择最具弹性的传感配置来实现。除了为睡眠研究提供一种新颖的、人工智能驱动的、可重构的工具设计之外，这项工作还将揭示新的多模态生物标志物，这些生物标志物可以在可穿戴设备上进行无创评估，用于检测睡眠阶段和障碍。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

A considerable amount of the population in the US and around the world suffers from a chronic sleep disorder. However, the majority of these are not diagnosed or treated. There is a vital need for new wearable technologies to increase the capacity of sleep researchers to make further advances in investigating sleep, understanding sleep pathologies, and to improve the ability of clinicians to reliably detect and treat sleep disorders. This award investigates the use of an artificial intelligence driven, reconfigurable sleep monitoring system to transform sleep research in the clinic and at-home. A sensor fusion strategy backed by artificial intelligence to ultra-miniaturize the sleep assessment instruments and explore novel sleep-related biomarker features have the transformative potential to invigorate sleep research for more efficient and accurate diagnosis and treatment of sleep disorders. Combining lower cost with better ergonomic comfort, and more efficient data analysis will pave the way for rapid translation, adoption, and effective deployment of these technologies for home-use in real-world settings. The research results from this award have the potential to positively influence the continuous monitoring instrumentation required for other chronic conditions such as heart diseases. This project enables several motivating opportunities for outreach and education including the use of technologies to interface sleep, learning new artificial intelligence and data analytics skills, analyzing sleep performance to unlock its mysteries, and the impact of sleep in the wellness and efficiency of society. Several such activities are planned with the products of this research project to reach out to younger generations, educators, other researchers in the field and public-at-large.This award integrates two parallel efforts combining innovations in hardware and data analytics: 1) enabling an adaptable and reconfigurable embedded system platform in the form factors of an adhesive patch, and 2) developing state-of-the-art machine learning techniques incorporating the data-driven models necessary for improving sleep monitoring system resilience. The hardware system fuses multimodal wearable sensors, combining Near Infrared Spectroscopy (NIRS) with other traditional sleep related signal sensors, in skin conformable substrates, to collect data on multiple body locations. The efficacy of the system will be assessed in terms of improving conformability and flexibility, and reducing the system real estate and cost. The data analytics platform includes 1) signal processing to enable data-driven metrics for signal quality assessment for a given inference task, 2) inference models based on transfer learning techniques and diverse datasets for detection of sleep events and disorders using new sensing modalities, and 3) Bayesian Neural Network supported sensor selection for improving the resilience and adaptability of sleep sensor systems. The `adaptation' will take place by selecting the most resilient sensing configurations over design iterations and in real-time during operation. In addition to allowing a novel, artificial intelligence-driven and reconfigurable tool design for sleep research, this effort will also shed light into novel multimodal biomarkers assessed noninvasively in wearable form factors for detection of sleep stages and disorders.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.