基于深度学习的手功能主动康复系统及其改善运动功能神经机制的研究

手部运动功能的恢复一直是脑卒中后康复治疗的焦点和难点。本项目拟针对脑卒中患者手功能康复的治疗需求，研究能够提高单手多类动作模式识别准确率的脑机接口算法，进而研发康复训练系统改善手功能康复治疗效果，并通过临床实验阐明与手功能主动康复相关的脑神经机制。第一，提出了两种脑机接口算法，包括脑肌电深度学习融合算法和基于最优多元经验模态分解的共空域模式算法，并通过临床实验来验证其有效性。第二，在课题组前期研发设备的基础上，结合上述算法研究多类动作模式手功能主动康复训练系统。第三，利用脑电微状态、神经血管耦合、脑电近红外光谱融合和多模态脑连接及脑网络分析，揭示手功能主动康复训练时大脑功能状态的变化规律。预计该课题的研究成果能够为手功能康复临床治疗提供新的思路和理论基础。项目申请人已具有肌电动作模式分类、脑电神经成像、脑连接及脑网络分析的研究经历，并已获得脑机接口控制系统和所提出算法的初步研究结果。

The recovery of hand motor function has always been the focus and difficulty of post-stroke rehabilitation therapy. Aiming at the treatment demands of hand functional rehabilitation of stroke patients, this project will investigate the brain-computer interface (BCI) algorithm which can improve the accuracy of single hand multi movement pattern recognition, and then develop a rehabilitation training system to improve the treatment effect of hand function rehabilitation, and clarify the brain nerve mechanism related to the active rehabilitation of hand function through clinical experiments. Firstly, two kinds of BCI algorithms will be proposed, including electroencephalography-electromyography (EEG-EMG) deep learning fusion algorithm and the common spatial pattern algorithm based on the optimal multiple empirical mode decomposition. The effectiveness of these algorithms will be validated through clinical experimental experiments. Secondly, on the basis of the equipment developed earlier by our research group, an active rehabilitation training system of hand function using multi-movement patterns will be studied by combining the above BCI algorithms. Thirdly, the changes of brain functional state during hand function active rehabilitation training will be revealed by the EEG microstates, neurovascular coupling, the fusion of EEG and functional near-infrared spectrum (fNIRs), multimodal brain connectivity, and brain networks analysis. It is expected that the research results of this project can provide new ideas and theoretical basis for clinical treatment of hand functional rehabilitation. The project applicant already has the research experiences about EMG-based movement classification, EEG neuroimaging, brain connectivity, and brain network analysis. Moreover, the project applicant has obtained the preliminary research results of the BCI control system and the proposed algorithm.