CAREER: Robust Decoding of Neural Command for Real Time Human Machine Interactions

职业：实时人机交互的神经命令的鲁棒解码

基本信息

批准号：
1847319
负责人：
Xiaogang Hu
金额：
$ 54.95万
依托单位：
University of North Carolina at Chapel Hill
依托单位国家：
美国
项目类别：
Continuing Grant
财政年份：
2019
资助国家：
美国
起止时间：
2019-04-01 至 2023-03-31
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1847319&HistoricalAwards=false
关键词：
CAREER Robust Decoding Neural Command

项目摘要

The human hand can produce complex dexterous movements, unmatched by any current robotic hand. Such sophisticated movements are often taken for granted. A majority of individuals with a stroke, however, tend to have persistent hand functional deficits, limiting their ability of living independently. Human-machine interactions hold great potential to restore motor functions of stroke survivors. Recently advanced rehabilitative or assistive techniques (e.g., hand exoskeletons) have the ability to substantially enhance motor functions. However, few of these state-of-the-art techniques have been successfully translated to end users, and one critical limiting factor is the challenge in controlling the many movement directions robustly. Therefore, there is an urgent need to develop non-invasive and robust neural decoding approaches for human-machine interactions that can directly translate to clinical applications. Accordingly, this project aims to decode the neural command sent from the brain that controls individual finger movements. This is accomplished by reading activities in the spinal cord using muscle electrical signals obtained from the skin surface. The decoded finger-specific neural command can then be used to control rehabilitation or assistive robots, which can substantially enhance the quality of human-machine interactions. This approach can also facilitate wide applications of robotic rehabilitation or assistance in stroke survivors. The non-invasive nature of the techniques has a great potential for readily clinical translations. The proposed research will be integrated with education through graduate and undergraduate research involvement and new course development. Summer projects and demonstration materials on human-machine interactions will be developed for K-12 students. Outreach programs will be organized to expose the proposed research topics to underrepresented students, highlight the opportunities in science and engineering, and promote students interests in choosing future STEM careers.The principal investigator's long-term research goal is to develop highly innovative non-invasive tools for human-machine interactions, with a particular interest in better understanding the neuromechanical properties of the upper extremity, and improve the functional performance in individuals with a central or peripheral injury. Toward this goal, this project aims to decode the descending neural command that controls individual finger movements by extracting spinal motoneuron discharge activities using source separation of high-density electromyogram signals (HD-EMG) from finger muscles. The non-invasive, robust, and real-time neural decoding technique developed will be easy to implement, can accommodate the different impairment levels of individual stroke survivors, and will substantially improve the control quality of exoskeleton or neuroprosthesis. The Research Plan is organized under three aims. The FIRST AIM is to develop non-invasive offline and real-time neural decoding approaches based on spinal motoneuron discharge probabilities at the population level that are directed at a designated finger. This aim addresses the need for non-invasive human-machine interface signals that allow robust and intuitive interaction between humans and machines. Surface EMG signals will be recorded over the targeted extrinsic muscles using an 8x16 channel electrode array with an inter-electrodedistance of 10 mm. Motoneuron discharge activities will be obtained from different independent component analysis (ICA)-based HD EMG decomposition methods that will be evaluated on both simulated and experimental EMG data obtained from stroke survivors and healthy control subjects. The decoding accuracy will be evaluated by comparing the decoded neural drive with finger force output and joint angles. Given that binary motoneuron discharge events are used, the decoded neural drive signals are expected to be robust to changes in action potential properties in the EMG signals, background noise, and motion artifacts. The evaluation of the performance and boundary conditions of different source separation algorithms can further ensure robust decoding performance in a variety of situations, especially in clinical populations. The SECOND AIM is to classify the neural command specific to individual finger movements. This aim addresses the need for effective control of individual/flexible finger movement in developing human-machine interactions. Surface EMG signals will be recorded over the extrinsic forearm muscles using an 8x16 channel HD EMG electrode array and over the intrinsic extensors muscles to fingers using an 8x4 channel grid. Different features from HD EMG activities and from motor unit (MU) distributions will be extracted. With macro and micro level features, different muscle activation regions will be identified for individual fingers using pattern classification approaches. The neural drive associated with specific finger movement will then be calculated based on MU discharge activities of a specific finger. The classified neural command signals can enable robust and flexible control of individual finger movements non-invasively, and dramatically enhance the dexterity of hand function in clinical populations. The THIRD AIM is to quantify the performance of the decoding technique by controlling a non-invasive neuroprosthesis for dexterous finger grasp patterns. A transcutaneous nerve stimulation technique developed in the PI's group will be used to elicit flexible individual and coordinated finger movements. The neural stimulation system targeting the affected hand of stroke survivors will be controlled by the decoded neural drive from the contralateral/unaffected arm (particularly if the stroke is severe) or from the affected arm, with time-sharing between stimulations and recordings. The force output (force absolute error and force variability) of neural drive controlled stimulation will be compared with the global EMG controlled stimulation to evaluate the performance of the neural decoding technique. The overall outcomes of the project are expected to ultimately allow stroke survivors to intuitively interact with rehabilitative/assistive devices in a robust and non-invasive manner.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.

人手可以产生复杂的灵巧动作，这是目前任何机器人手都无法比拟的。如此复杂的动作通常被认为是理所当然的。然而，大多数中风患者往往存在持续的手部功能缺陷，限制了他们独立生活的能力。人机交互对于恢复中风幸存者的运动功能具有巨大的潜力。最近先进的康复或辅助技术（例如手外骨骼）能够显着增强运动功能。然而，这些最先进的技术很少被成功地转化为最终用户，并且一个关键的限制因素是稳健控制许多运动方向的挑战。因此，迫切需要开发可直接转化为临床应用的非侵入性、鲁棒性的人机交互神经解码方法。因此，该项目旨在解码从大脑发送的控制单个手指运动的神经命令。这是通过使用从皮肤表面获得的肌肉电信号读取脊髓中的活动来实现的。解码后的手指特定神经命令可用于控制康复或辅助机器人，从而大大提高人机交互的质量。这种方法还可以促进机器人康复或中风幸存者援助的广泛应用。该技术的非侵入性性质具有易于临床转化的巨大潜力。拟议的研究将通过研究生和本科生的研究参与和新课程开发与教育相结合。将为 K-12 学生开发有关人机交互的夏季项目和演示材料。将组织外展计划，向代表性不足的学生展示拟议的研究课题，突出科学和工程领域的机会，并提高学生选择未来 STEM 职业的兴趣。首席研究员的长期研究目标是开发高度创新的非侵入性工具用于人机交互，特别感兴趣的是更好地了解上肢的神经力学特性，并提高中枢或外周损伤个体的功能表现。为了实现这一目标，该项目旨在通过使用来自手指肌肉的高密度肌电信号（HD-EMG）的源分离提取脊髓运动神经元放电活动来解码控制个体手指运动的下行神经命令。所开发的非侵入性、鲁棒性和实时神经解码技术将易于实施，可以适应个体中风幸存者的不同损伤程度，并将显着提高外骨骼或神经假体的控制质量。该研究计划根据三个目标进行组织。第一个目标是开发基于针对指定手指的群体水平的脊髓运动神经元放电概率的非侵入性离线和实时神经解码方法。这一目标满足了对非侵入式人机界面信号的需求，从而允许人与机器之间进行稳健且直观的交互。将使用电极间距离为 10 mm 的 8x16 通道电极阵列记录目标外在肌肉上的表面 EMG 信号。运动神经元放电活动将通过不同的基于独立成分分析 (ICA) 的 HD EMG 分解方法获得，这些方法将根据从中风幸存者和健康对照受试者获得的模拟和实验 EMG 数据进行评估。将通过将解码的神经驱动与手指力输出和关节角度进行比较来评估解码准确性。鉴于使用二元运动神经元放电事件，解码的神经驱动信号预计对 EMG 信号、背景噪声和运动伪影中动作电位特性的变化具有鲁棒性。对不同源分离算法的性能和边界条件的评估可以进一步确保在各种情况下，特别是在临床人群中的鲁棒解码性能。第二个目标是对特定于单个手指运动的神经命令进行分类。这一目标解决了在开发人机交互时有效控制单独/灵活的手指运动的需求。将使用 8x16 通道 HD EMG 电极阵列记录外在前臂肌肉上的表面 EMG 信号，并使用 8x4 通道网格记录手指内在伸肌上的表面 EMG 信号。将提取 HD EMG 活动和运动单位 (MU) 分布的不同特征。通过宏观和微观层面的特征，将使用模式分类方法识别各个手指的不同肌肉激活区域。然后，将根据特定手指的 MU 放电活动来计算与特定手指运动相关的神经驱动。分类后的神经命令信号可以实现对个体手指运动的非侵入性鲁棒、灵活控制，并显着增强临床人群手部功能的灵活性。第三个目标是通过控制非侵入性神经假体以实现灵巧的手指抓握模式来量化解码技术的性能。 PI 小组开发的经皮神经刺激技术将用于引发灵活的个体和协调的手指运动。针对中风幸存者受影响手的神经刺激系统将由来自对侧/未受影响手臂（特别是中风严重时）或受影响手臂的解码神经驱动控制，刺激和记录之间分时进行。神经驱动控制刺激的力输出（力绝对误差和力变异性）将与全局肌电图控制刺激进行比较，以评估神经解码技术的性能。该项目的总体成果预计最终将允许中风幸存者以稳健和非侵入性的方式直观地与康复/辅助设备互动。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力优势进行评估，被认为值得支持以及更广泛的影响审查标准。

项目成果

期刊论文数量（23）

专著数量（0）

科研奖励数量（0）

会议论文数量（0）

专利数量（0）

Virtual Reality for Evaluating Prosthetic Hand Control Strategies: A Preliminary Report

用于评估假手控制策略的虚拟现实：初步报告

DOI：
10.1109/embc46164.2021.9630555
发表时间：
2021-11
期刊：
2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC
影响因子：
0
作者：
Xie, Jason;Hu, Xiaogang
通讯作者：
Hu, Xiaogang

Concurrent Estimation of Finger Flexion and Extension Forces using Motoneuron Discharge Information

使用运动神经元放电信息同时估计手指屈曲和伸展力

DOI：
10.1109/tbme.2021.3056930
发表时间：
2021-04
期刊：
IEEE Transactions on Biomedical Engineering
影响因子：
4.6
作者：
Zheng, Yang;Hu, Xiaogang
通讯作者：
Hu, Xiaogang

Real-time isometric finger extension force estimation based on motor unit discharge information

DOI：
10.1088/1741-2552/ab2c55
发表时间：
2019-10-10
期刊：
Journal of Neural Engineering
影响因子：
4
作者：
Yang Zheng;Xiaogang Hu
通讯作者：
Xiaogang Hu

Elicited Finger and Wrist Extension Through Transcutaneous Radial Nerve Stimulation

通过经皮桡神经刺激引起手指和手腕伸展

DOI：
10.1109/tnsre.2019.2930669
发表时间：
2019-07-24
期刊：
IEEE Transactions on Neural Systems and Rehabilitation Engineering
影响因子：
4.9
作者：
Yang Zheng;Xiaogang Hu
通讯作者：
Xiaogang Hu

Real-time finger force prediction via parallel convolutional neural networks: a preliminary study

通过并行卷积神经网络进行实时手指力预测：初步研究

DOI：
10.1109/embc44109.2020.9175390
发表时间：
2020-07
期刊：
Proceedings of IEEE Engineering in Medicine and Biology Society Annual Meeting
影响因子：
0
作者：
Xu, Feng;Zheng, Yang;Hu, Xiaogang
通讯作者：
Hu, Xiaogang