脑电信号中肌电噪声去除的新探索

Electroencephalography (EEG) reflects the weak spontaneous and rhythmic electrical activity within the neurons of the brain. It has been widely used for medical diagnosis in clinic, health care and other fields. However, there exist a number of muscles spatially and widely distributed across the human head. Electromyography (EMG) arising from the muscles would have a serious interference for EEG and its subsequent analysis. Due to the complexity of spatial, temporal and spectral characteristics of EMG, EMG removal currently becomes a difficult problem in the research field of EEG. Through our profound analysis for existing difficulties, this proposal aims at a new exploration of this problem by developing novel methods based on joint blind source separation (JBSS), a popular technique proposed in recent years. On one hand, given that EEG and EMG sources and mixing matrices may change over time, dynamic JBSS will be developed to more accurately recover EMG sources; on the other hand, considering that the number of EEG and EMG sources is generally greater than that of channels, underdetermined JBSS will be proposed to maximally separate EMG sources. Besides, a benchmark database and a software toolbox will be provided to the public. A set of effective solutions will be proposed for the problem of EMG removal from EEG. The launching of this study will significantly improve the quality of EEG acquired in medical diagnosis and mobile health care. It is of great importance to further study the underlying electrical activity of the brain.

脑电信号反映了脑部神经细胞群微弱的自发性、节律性电生理活动，已被广泛应用于医学临床诊断及健康监护等领域。然而，人体头部肌肉分布较多较广，肌电会对脑电产生严重干扰而影响后续分析。由于肌电的时空频域特性复杂，使其去除成为当前脑电研究领域的难题。本项目拟结合近年来兴起的联合盲源分离（JBSS)技术，通过对难点的深刻剖析，设计有针对性的方案，对该难题进行新的探索：一方面，针对脑电和肌电信号源及混合矩阵随时间变化的情况，设计动态JBSS方法，更加准确地估计肌电信号源；另一方面，针对脑电和肌电信号源数之和大于通道数的情况，设计欠定JBSS方法，最大限度地分离肌电信号源。此外，本项目将建立开放共享的基准数据库和算法工具箱，提出一套完善的脑电信号中肌电噪声去除的解决方案。本研究的开展将有助于提升医学临床诊断及移动健康监护中脑电的获取质量，对进一步研究大脑真实的电生理活动具有重要意义。

脑电信号反映了脑部神经细胞群微弱的自发性、节律性电生理活动，已被广泛应用于医学临床诊断及健康监护等领域。然而，人体头部肌肉分布较多较广，肌电会对脑电产生严重干扰而影响后续分析。由于肌电的时空频域特性复杂，使其去除成为当前脑电研究领域的难题。本项目围绕该难题，针对单通道、少数通道和多通道三种情况提出了一系列基于联合盲源分离技术的降噪新方案，并且针对该问题探索了单/少数通道与多通道之间的相互关系，所提出的单通道EEMD-JBSS、少数通道MEMD-JBSS和多通道IVA方法在领域内均具有很强的创新性，相关工作已被国际知名机构数十位院士和Fellow的研究团队引用，相关脑电降噪算法已被包括美国加州大学圣迭戈分校和佛罗里达大学两个著名团队进行具体应用，多项研究成果进入了（扩展版）ESI高被引。项目组设计了开源共享的软件工具箱ReMAE，集成了当前state-of-the-art的方法，相关成果将对人机交互、脑科学、主动健康和康复工程等民用领域的研究起到积极作用。项目共发表了高水平论文21篇（包括IEEE期刊12篇），申请国家发明专利5项（授权1项），获得软件著作权1项，指导研究生8名，与国内外著名研究组进行了学术交流与合作。