Closed-Loop Stimulus Optimization to Increase Communication Efficiency in Brain-Computer Interfaces

摘要 这一行政补充是为了回应特别利益通知，以改善人工 智能和机器学习（AI/ML）-NIH支持的sata的就绪性（NOT-OD-21-094）。总结 家长奖。增强和替代性通信（AAC）系统由以下人员使用： 沟通和运动障碍，如肌萎缩性侧索硬化症（ALS），沟通和 与环境互动。存在由访问方法控制的常规AAC设备 例如触摸、切换、头部跟踪和眼睛注视;然而，这些访问方法变得困难或 当持续的肌肉控制或随意运动控制丧失时，不可能使用。有脑机 脑机接口（BCI）通信系统，如P300拼写器，使用感官刺激来引出和 然后检测脑电图（EEG）数据中的感觉神经反应。然而，与 刺激驱动的BCI是次优的，这是由于依赖于固有的噪声EEG数据和高度可变的神经元 BCI控制的响应。BCI通信速率可以通过利用 实时EEG数据，以优化BCI系统参数，从而在以下条件下最大限度地提高BCI性能 不确定性的条件。这项工作研究了一种新的闭环刺激选择算法， P300拼写者的刺激呈现时间表基于所测量的EEG数据和 BCI系统对用户意图的信念。目的1开发和测试新的算法在一个队列的残疾人- 身体的个人来评估闭环刺激选择的实时可行性和实用性。目标2将 在患有ALS的个体的队列中测试闭环刺激选择算法以评估性能 目标BCI最终用户中的算法。本补充的目的。目前，对大型、 不同的BCI数据集，包括BCI算法开发的目标BCI最终用户，特别是 数据饥渴型深度学习模型的流行。基于NIH支持的10多年研究，我们 从P300拼写研究中获得了大量的单次和多次数据， 参与者和ALS参与者使用不同的刺激呈现范式。以公平为导向 原则，在本补充：（1）我们将执行数据策展，数据清洗和数据工程，以开发 具有通用数据和元数据元素的跨平台可读P300拼写器数据集， 公开可用的数据集;我们将在（2）AI/ML应用程序中演示此数据集的可用性 专注于开发强大的数据表示，以减轻EEG数据中可变性的负面影响 AI/ML算法;以及（3）专注于数据科学技能发展的学生研究计划， AI/ML。一个大型的、包容性的、可访问的BCI数据集将对BCI社区和整个世界产生重大影响。 更广泛的AI/ML社区，因为它将支持开发和比较新数据表示的研究， 刺激范例和神经信号解码算法，以建立BCI作为可行的AAC设备。

ABSTRACT This administrative supplement is in response to the Notice of Special Interest to improve the artificial intelligence and machine learning (AI/ML)-Readiness of NIH-supported sata (NOT-OD-21-094). Summary of Parent Award. Augmentative and alternative communication (AAC) systems are used by people with communication and motor disabilities, such as amyotrophic lateral sclerosis (ALS), to communicate and interact with their environment. There are conventional AAC devices that are controlled by access methods such as touch, switch, head tracking and eye gaze; however, these access methods become difficult or impossible to use when sustained muscle control or voluntary motor control is lost. There are brain-computer interface (BCI) communication systems, such as the P300 speller, that use sensory stimulation to elicit and then detect sensory neural responses in electroencephalography (EEG) data. However, communication with stimulus driven BCIs is suboptimal due to relying on inherently noisy EEG data and highly variable neural responses for BCI control. BCI communication rates can potentially be improved by leveraging information in EEG data in real-time to optimally tune the BCI system’s parameters to maximise BCI performance under conditions of uncertainty. This work investigates a novel closed-loop stimulus selection algorithm that optimises the stimulus presentation schedule of the P300 speller in real-time based on the measured EEG data and the BCI system’s belief about the user’s intent. Aim 1 develops and tests the novel algorithm in a cohort of abled- bodied individuals to evaluate the real-time feasibility and utility of closed-loop stimulus selection. Aim 2 will test the closed-loop stimulus selection algorithm in a cohort of individuals with ALS to assess the performance of the algorithm in target BCI end users. Goals of this Supplement. There is a current unmet need for large, diverse BCI datasets that include target BCI end users for BCI algorithm development, particularly with the popularity of data hungry deep learning models. Based on NIH-supported research for 10+ years, we have acquired a large amount of single- and multi-session data from P300 speller studies with abled-bodied participants and participants with ALS using different stimulus presentation paradigms. Guided by FAIR principles, in this supplement: (1) we will perform data curation, data cleaning and data engineering to develop a cross-platform readable P300 speller dataset with common data and metadata elements and make this dataset publicly available; and we will demonstrate the usability of this dataset in (2) an AI/ML application focused on developing robust data representations to mitigate the negative effect of variabilities in EEG data on AI/ML algorithms; and (3) in student research programs focused on skill development in data science and AI/ML. A large, inclusive and accessible BCI dataset will have significant impact in the BCI community and the broader AI/ML community, as it will support research to develop and compare novel data representations, stimulus paradigms and neural signal decoding algorithms towards establishing BCIs as viable AAC devices.