Brain-Computer Interface in dynamic tasks with deep learning and functional connectivity analysis

具有深度学习和功能连接分析的动态任务中的脑机接口

• 批准号: 10292336
• 负责人: Hubert Charles Cecotti
• 金额: $ 39.67万
• 依托单位: CALIFORNIA STATE UNIVERSITY FRESNO
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-09-01 至 2024-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10292336
• 关键词: Active Learning; Applied Research; Architecture; Area; Asian Americans; Attention; Biomedical Engineering; Brain; Calibration; California; Characteristics; Clinical; Communication; Computers; Craniocerebral Trauma; Data; Data Science; Data Set; Detection; Dimensions; Disabled Persons; Drops; Education; Electroencephalogram; Electroencephalography; Environment; Event; Event-Related Potentials; Face; Female; Goals; Hand; Hispanic-serving Institution; Image; Institution; Interdisciplinary Study; Knowledge; Life; Machine Learning; Mental concentration; Methodology; Methods; Modeling; Monitor; Native Americans; Neural Network Simulation; Neuromuscular Diseases; Pacific Island Americans; Performance; Physiological; Probability; Process; Property; Psychology; ROC Curve; Research; Research Infrastructure; Research Project Grants; Series; Signal Transduction; Societies; Source; Stroke; Students; System; Techniques; Technology; Time; Time Series Analysis; Translating; Underrepresented Minority; Universities; Update; Visual; base; brain computer interface; clinical application; computer science; connectome; convolutional neural network; deep learning; expectation; experience; experimental study; improved; information processing; innovation; mental state; nervous system disorder; primary outcome; public health relevance; recurrent neural network; rehabilitation strategy; relating to nervous system; response; robot control; signal processing; skills; success; tool; virtual; visual information

Abstract The PI proposes a high-impact multi-disciplinary research project to develop and validate machine learning al- gorithms for shift-detection in electroencephalogram (EEG) signals with applications to brain-computer interface to make them more reliable. Brain-computer interface is a means of communication for severely disabled peo- ple by decoding brain responses and translating their detection into commands with applications such a virtual keyboard or robotic control systems. Current brain-computer interface systems cannot be efficiently deployed in clinical setting due to their inability to properly take into account the non-stationarity properties of the evoked brain responses in the electroencephalogram signal. This project aims at enhancing the brain decoding perfor- mance when the task changes over time. The PI proposes to investigate the effects of well defined types of data shifts: covariate shift, probability shift, and concept shift to enhance brain decoding performance in changing tasks. The goals of this proposal are: 1) to characterize in event related potential (ERP) components neural signatures corresponding to task changes by using EEG recordings and machine learning techniques for single- trial detection. 2) to research in functional brain connectivity neural signature corresponding to task changes by using EEG recordings and directed model-based and model free techniques of functional brain connectivity. 3) to combine and adapt machine learning techniques to detect when changes occur during a task. This proposal will significantly improve the infrastructure of research and education at California State University Fresno, Hispanic- Serving Institution and an Asian American and Native American Pacific Islander-Serving Institution, introducing biomedical engineering research experiences to underrepresented minority and female students in computer science and psychology students. This would allow them to experience different stages of the scientific method, and acquire fundamental skills related to data science applied to physiological signals with potential impact on society for improving the life of severely disabled people.

抽象的 PI 提出了一个高影响力的多学科研究项目来开发和验证机器学习算法 脑电图 (EEG) 信号偏移检测算法及其在脑机接口中的应用 使他们更加可靠。脑机接口是严重残障人士的一种沟通手段 可以通过解码大脑反应并将其检测结果转化为应用程序（例如虚拟应用程序）的命令来实现这一点。 键盘或机器人控制系统。当前的脑机接口系统无法有效部署 在临床环境中，由于他们无法正确考虑诱发的非平稳特性 大脑对脑电图信号的反应。该项目旨在增强大脑解码性能 当任务随时间变化时。 PI 建议调查明确定义的数据类型的影响 转变：协变量转变、概率转变和概念转变，以增强大脑在变化中的解码性能 任务。该提案的目标是：1）表征事件相关电位（ERP）神经成分 通过使用脑电图记录和机器学习技术来识别与任务变化相对应的签名 试探检测。 2）研究与任务变化相对应的功能性大脑连接神经特征 使用脑电图记录以及基于模型和无模型的功能性大脑连接的定向技术。 3）到 结合并调整机器学习技术来检测任务期间何时发生变化。该提案将 显着改善加州州立大学弗雷斯诺分校的研究和教育基础设施 服务机构和亚裔美国人和美洲原住民太平洋岛民服务机构，介绍 为计算机领域代表性不足的少数族裔和女学生提供生物医学工程研究经验 科学和心理学的学生。这将使他们体验科学方法的不同阶段， 并获得与应用于生理信号的数据科学相关的基本技能，这些技能对 改善严重残疾人生活的社会。