SCH: Enhanced detection of impending problem behavior in people with intellectual and developmental disabilities through multimodal sensing and machine learning

患有智力和发育障碍(IDD)的儿童表现出“问题行为”的风险增加，这会使他们面临受伤、离开教室或住院的风险。在美国，大约六分之一的儿童和青少年被诊断出患有IDD，其中一半人经历了某种形式的问题行为。接受过应用行为分析(ABA)培训的治疗师可以帮助确定问题行为发生的原因以及如何预防。这些治疗师观察儿童，试图通过改变儿童的环境来唤起问题行为，然后尝试可能改变行为的事情，并观察行为数据是否发生变化。因为问题行为可能会在这个过程中被触发，所以这种策略有时会将他们或他们的患者置于危险之中。这也需要很多时间。可穿戴技术和先进的计算策略可以帮助提高预防问题行为的策略的安全性和有效性。具体地说，穿戴在衣服或手腕上的小型传感器可以提供关于孩子身体的数据，或者像心率或汗水这样的“生理反应”。然后，机器学习可以用来确定身体信号的组合意味着问题行为即将发生。这个项目分为两个阶段。在第一阶段，研究小组将设计新的传感器来检测生物信号，如出汗、运动和心率。然后，研究小组将测量这些传感器的工作情况。这包括询问IDD患者他们对传感器的看法。在此基础上，该团队将更换传感器，然后在一项更大的研究中使用它们。目标是测试该系统是否可以预测问题行为，当它在现实世界中与真实的治疗师一起使用时，它的效果如何，以及用户对该系统的看法。这项研究的结果将帮助研究人员和从业者了解这种可穿戴技术作为支持有问题行为和IDD的人的一部分是否有帮助和是否被接受。该项目建议整合尖端可穿戴传感、情感计算、机器学习以及行为和临床科学方面的跨学科专业知识，以增强和改变现有的针对IDD儿童和青少年问题行为的行为干预模式。问题行为，包括自我伤害、攻击、财产破坏和走失，不仅会导致重伤或死亡，还会干扰参与学校、家庭和其他社区环境的能力。在问题行为和IDD的背景下，本项目将从根本上推进基于多模式可穿戴传感的预测机器学习模型设计的科学和技术方法论。这两个研究方向是：(1)多模式传感器框架设计；(2)实时前兆预报。通过这些努力，该项目将在以下方面取得根本性的科学和技术进步：(I)低功耗、开放访问、以用户为中心的可穿戴传感器框架，可以感知用于情感计算的生理反应和手势；以及(Ii)一套新颖的、基于临床的半监督机器学习模型，用于预测问题行为，可供行为干预者实时使用。这项研究与该领域现有工作的一个重要创新之处在于，该团队提议通过问题行为的前兆而不是行为本身来解决问题行为的检测，目的是提高会话的安全性和效率。这些科学和技术进步将在最先进的临床和行为科学框架内创造。拟议的工作将促进工程和健康科学的跨学科研究。该团队提出了一些将在STEM教育中产生更广泛影响的推广和教育活动：i)通过第一自闭症与创新中心的神经多样性团队让ASD患者直接参与研究；ii)为早期临床科学家提供跨学科培训机会；iii)为高中、本科生和研究生提供研究机会；iv)为高中教师提供研究机会；v)将研究带入课堂；vi)通过研讨会、演讲和出版传播研究。重点是从少数群体和代表性不足的群体中招募候选人，以提高STEM的多样性。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Children with Intellectual and Developmental Disabilities (IDD) are at increased risk of showing “problem behavior” that place them at risk of getting hurt, removed from the classroom, or hospitalized. Approximately 1 in 6 children and adolescents in the United States are diagnosed with IDD and half of them experience some form of problem behavior. Therapists trained in Applied Behavior Analysis, or ABA, can help determine why problem behavior happens and how to prevent it. These therapists watch children, try to evoke problem behaviors by changing a child’s environment, then try things that might change behavior, and see if the behavioral data changes. Because problem behavior can be triggered during this process, this strategy sometimes put them or their patient at risk. It also takes a lot of time. Wearable technology and advanced computational strategies could help increase the safety and helpfulness of strategies to prevent problem behavior. Specifically, small sensors worn in clothing or on the wrist could provide data about a child’s body, or “physiological responses,” like heart rate or sweat. Machine learning can then be used to determine what combination of body signals imply a problem behavior is about to happen. This project has two stages. In the first stage, the team will design new sensors that detect biological signals such as sweating, motion, and heart rate. The team will then measure how well these sensors work. This includes asking people with IDD what they think about the sensors. Based on that input, the team will change the sensors and then use them in a larger study. The goal is to test whether the system can predict problem behavior, how well it works when used in the real-world with real therapists, and what users think about the system. Results of this study will help researchers and practitioners understand if this kind of wearable technology is helpful and acceptable as part of supporting people with problem behavior and IDD. This project proposes to integrate transdisciplinary expertise in cutting-edge wearable sensing, affective computing, machine learning, and behavioral and clinical science to enhance and transform existing models of behavioral intervention for problem behaviors in children and adolescents with IDD. Problem behaviors, including self-injury, aggression, property destruction, and wandering not only can cause serious injury or death, but also interfere with the ability to participate in school, home, and other community settings. In the context of problem behavior and IDD, this project will fundamentally advance the scientific and the technological methodologies of multimodal wearable sensing-based design of predictive machine learning models. The two research thrusts are: (1) Design of multimodal sensor framework; and 2) Real-time precursor prediction. Across these thrusts, the project will make fundamental scientific and technological advancements in: (i) A low-power, open-access, user-centric wearable sensor framework that can sense physiological responses and gestures to be used for affective computing; and (ii) A set of novel, clinically grounded, semi-supervised machine learning models to predict problem behavior that can be used by behavioral interventionists in real-time. An important novelty of this research that separates it from existing work in the field is that the team proposes to address the detection of problem behavior through its precursors, rather than the behaviors themselves, with the goal of increasing the safety and efficiency of sessions. These scientific and technological advancements will be created within a state-of-the-art clinical and behavioral science framework. The proposed work will foster interdisciplinary research in engineering and health sciences. The team proposes a number of outreach and educational activities that will have broader impact in STEM education: i) involve individuals with ASD directly in the research through the Frist Center for Autism & Innovation’s Neurodiversity Corps; ii) provide interdisciplinary training opportunities for early stage clinical scientists; iii) provide research opportunity to high school, undergraduate, and graduate students; iv) provide research opportunity to high school teachers; v) bring research into the classroom; and vi) disseminate the research through seminars, presentation, and publication. Emphasis will be placed on recruiting candidates from minority and underrepresented groups to improve diversity in STEM.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.