Current Issue Cover


摘 要
脑-机接口(Brain-computer interface,BCI)系统通过采集、分析大脑信号,将其转换为输出指令,从而跨越外周神经系统,实现由大脑信号对外部设备的直接控制,进而用于替代、修复、增强、补充或改善中枢神经系统的正常输出。非侵入式脑-机接口由于具有安全性以及便携性等优点,得到了广泛关注和持续不断地研究。近年来,研究人员对脑信号编码方法的不断探索扩展了BCI系统的应用场景和适用范围;同时,脑信号解码方法的不断研发,极大地克服了脑电信号信噪比低的缺点,提高了系统性能,这都为构建高性能脑-机接口系统奠定了基础。本文综述了非侵入式脑-机接口编解码技术以及系统应用的最新研究进展,展望其未来发展前景,以期促进BCI系统的深入研究与广泛应用。
A survey on encoding and decoding technology of non-invasive brain-computer interface

Qiu Shuang,Yang Banghua,Chen Xiaogang,Wang Yijun,Xu Minpeng,Lu Bao-Liang,Gao Xiaorong,He Huiguang(Institute of Automation,Chinese Academy of Sciences)

A brain-computer interface (BCI) establishes a direct communication pathway between the living brain and an external device by acquiring and analyzing brain signals and converting them into output commands. This is used to replace, repair, augment, supplement or improve the normal output of the central nervous system. According to the placement position of the acquisition electrode, BCIs can be divided into invasive or noninvasive BCIs. An invasive BCI directly records or stimulates brain neurons through electrodes implanted into the brain by surgery. Although invasive BCIs record brain signals with a high signal-to-noise ratio, they are only used in animal experiments or patients with severe paralysis due to the high risk of surgery. Compared with invasive BCIs, noninvasive BCIs have been widely developed and used due to their safety and portability advantages. Elec-troencephalography (EEG) is the most commonly used brain signal for BCIs. The technologies used in EEG-based BCI systems mainly include the coding methods used to generate brain signals and the decoding methods used to decode brain signals. In recent years, researchers" continuous exploration of coding methods has extended the application scenarios and applicability of the system; meanwhile, the continuous development of brain signal decoding methods has greatly overcome the shortcomings of the low signal-to-noise ratio of EEG signals and improved the system performance, which have laid the foundation for building high-performance BCIs. BCI systems can be divided into active, reactive or passive according to the coding methods. In an active BCI, a user can consciously control mental activities without relying on external stimuli. The motor imagery (MI)-based BCI system is a typical active BCI system based on frequency-specific changes in EEG signals, which can be modulated using mental rehearsal of a motor act without any motor output. In a reactive BCI, brain activity is triggered by an external stimulus, and the user reacts to the stimulus from the external world. A steady-state visual evoked potential (SSVEP)-based BCI and an event-related potential (ERP)-based BCI are the most widely studied. SSVEPs are brain responses that are elicited over the visual region when a user focuses on a flicker of a visual stimulus. An ERP-based BCI is usually based on a P300 component of ERP that is produced after the onset of the stimulus. A passive BCI provides the computer with the hidden state of the brain in the human-computer interaction process rather than control so that the computer can make ad-justments in time and realize humanized interaction. For example, affective BCIs and mental load BCIs. This report focuses on four main BCI systems, MI-based BCI, SSVEP-based BCI, ERP-based BCI, and affective BCI, and reviews their latest research progress in coding and decoding technology as well as the application of BCI systems. 1) MI-based BCI studies mostly focus on the classification of EEG patterns during MI tasks from different limbs, such as the left hand, right hand, and both feet. However, small instruction sets have difficulty satisfying the actual application requirements. Thus, fine MI tasks from the unilateral limb were proposed recently. This report focuses on reviewing the research progress of fine MI paradigms, including different joints of the unilateral limb, different directions of the unilateral limb, and different tasks of the unilateral limb. Decoding methods generally consist of two procedures: feature extraction and feature classification. Feature extraction extracts task-related and recognizable features from brain signals, and feature classification uses extracted features to distinguish the intentions of users. For MI decoding technology, this report first briefly introduces two-stage traditional decoding methods (e.g., common spatial pattern (CSP) and linear discriminant analysis (LDA)). After that, this report mainly summarizes the latest deep learning methods, which improve the decoding accuracy, and summarizes some migration learning methods, which reduce the need for calibration data. Finally, this report introduces the recent application of the MI-BCI system in the control of the mechanical arm and stroke rehabilitation. 2) SSVEP-based BCI systems have attracted wide attention due to their high in-formation transmission rate, strong stability and wide applicability. This report reviews recent research highlights of SSVEP-based BCI studies. For example, some new coding strategies have been developed to expand the instruction set or to improve users’ comfort, and some new decoding methods have been proposed to improve the performance of SSVEP detection. Additionally, this report summarizes the main application directions of SSVEP-based BCI systems, including communication, control and state monitoring, and reviews the latest application progress. 3) In ERP-based BCIs, the signal-to-noise ratio of the ERP response is relatively low, and the traditional rapid serial visual presentation paradigm aims to detect a single target. To overcome the above limitations, researchers have developed hybrid BCI paradigms by combining ERP and other paradigms, which have been summarized in this report. In this report, ERP decoding methods have been divided into four categories ac-cording to their functions: signal denoising methods, feature extraction methods, transfer learning methods and zero calibration algorithms. This report summarizes the recent progress of these decoding methods and ERP-based BCI applications. 4) Affec-tive BCI can recognize and modulate human emotion. This report summarizes the materials used to evoke emotion and new decoding methods, including multimodal fusion methods and transfer learning methods. Additionally, this report summarizes the application of affective BCI in healthcare. Finally, this report presents the future prospects of noninvasive BCI to promote in-depth research and future applications.