CAREER: Towards a Biologically Informed Intervention for Emotionally Dysregulated Adolescents and Adults with Autism Spectrum Disorder

研究人员计划改进依赖于虚拟和增强现实环境中使用的脑机接口的系统。这些改进将提高舒适性和可靠性时，使用的个人与自闭症谱系障碍。这些改善的好处将提高情绪调节和行为干预治疗的有效性。 人们越来越有兴趣用各种低成本和易于获得的基于技术的工具来补充这种行为临床治疗，以提高治疗效果。然而，研究表明，由于许多原因，通过现有的基于技术的自闭症谱系障碍（ASD）干预工具进行的培训通常不会推广到现实生活中的活动。研究人员将开发一种针对ASD的干预措施，以加强基于实时监测和分析的情绪调节策略。具体而言，计划中的脑电图（EEG）引导的脑机接口（BCI）技术可用于补充所有专注于情绪调节的临床治疗，以减少临床医生在每位患者身上花费的时间。新的科学发现和工程改进将为开发ASD干预技术做出巨大贡献，以（i）减少抑郁和焦虑;（ii）减少问题行为，包括社交互动中的攻击行为;以及（iii）减少不同环境中的功能障碍，包括学校，工作，家庭和社区。研究和教育目标将包括：（i）课程开发;（ii）将从K-12到研究生水平的研究人员纳入尖端的跨学科研究环境，以促进STEM职业;以及（iii）建立新的外展活动，向更广泛的公众通报拟议的研究成果和基于技术的ASD干预研究的最新技术进步。 这个特定项目的研究目标是引入一个框架，使EEG引导的闭环：（i）在基于技术的ASD干预期间监测个体的大脑反应，以及（ii）控制ASD个体情绪调节的临床治疗策略提示的呈现。特别地，对于这样的人机接口：（1）在ASD干预期间通过EEG监测大脑反应的提议是新颖的，以及（2）制定用于基于模型的最佳EEG引导的闭环临床治疗策略提示呈现的设计原则是变革性的。 EEG的实时概率分析的这些主张是独特的，并且提出了一个潜在的改变游戏规则的机会，以提高现有的基于技术的ASD干预情绪调节的泛化效果。 该项目还将有助于开发新的机器学习算法和神经科学方法，以识别与情绪调节相关的EEG特征，从而在痛苦和非痛苦条件之间进行分类，并区分不同的痛苦水平。 开发的模型将基于一个坚实的数学框架，基于变分自动编码器和贝叶斯最优统计推断，信息理论措施的功能选择有效的学习，和计算效率的优化模块化和子模块单调或非单调功能。 优化算法将提供计算效率高的解决方案，产生（次）最佳的功能选择策略与性能保证。这个奖项反映了NSF的法定使命，并已被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

The investigator plans to improve systems that rely on brain-computer interfaces that are used in virtual and augmented reality environments. These improvements will enhance comfort and reliability when used by individuals with disorders within the autism spectrum. The benefits of these improvements will advance the effectiveness of treatments for emotion regulation and behavioral interventions. There is a growing interest in complementing such behavioral clinical treatments with various low-cost and easy-to-access technology-based tools to improve therapy efficacy. However, it was shown that training through existing technology-based Autism Spectrum Disorder (ASD) intervention tools does not usually generalize to real-life activities for many reasons. The investigator will develop an intervention for ASD to reinforce emotion regulation strategies based on real-time monitoring and analysis. Specifically, the planned electroencephalography (EEG)-guided brain-computer interface (BCI) technology could be used to complement all clinical treatments that focus on emotion regulation to decrease clinician time spent with each patient. The novel scientific discoveries and engineering enhancements will have overreaching contributions to develop ASD intervention techniques for (i) decreased depression and anxiety; (ii) decreased problematic behaviors including aggression in social interactions; and (iii) decreased functional impairment across different settings including school, work, home and community. Research and education goals will include: (i) course development; (ii) inclusion of researchers from K-12 to graduate level in cutting-edge interdisciplinary research environment to promote STEM careers; and (iii) establishing new outreach activities to inform the broader public about the proposed research outcomes and the latest technological advancements in research for technology-based ASD intervention. The research objective of this specific project is to introduce a framework that will enable EEG-guided closed-loop: (i) monitoring of the brain responses of individuals during technology-based ASD intervention, and (ii) control of the presentation of clinical treatment strategy cues for emotion regulation in individuals with ASD. In particular, for such human-computer interfaces: (1) the proposition of monitoring brain responses through EEG during ASD intervention is novel, and (2) formulating design principles for model-based optimal EEG-guided closed-loop clinical treatment strategy cue presentation is transformative. These propositions of real-time probabilistic analysis of EEG are unique, and present a potentially game-changing opportunity to advance the generalization effect of existing technology-based ASD intervention for emotion regulation. This project will also contribute to developing new machine learning algorithms and neuroscience methods to identify EEG features associated with emotion regulation to classify between distress and non-distress conditions, and to distinguish among different distress levels. The developed models will be based on a solid mathematical framework based on variational autoencoders and Bayesian optimal statistical inference, information theoretic measures of feature selection for efficient learning, and computationally efficient optimization of modular and submodular monotonic or non-monotonic functions. Optimization algorithms will provide computationally efficient solutions that generate(sub)optimal feature selection strategies with performance guarantees.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.