关键词: 临床试验, 深度学习, 诊断, 脑电图范式, 电极, 运动成像, 康复, 传感器, 稳态视觉诱发电位, 跨学科学习

Abstract:

Electroencephalography-based brain–computer interfaces have revolutionized the integration of neural signals with technological systems, offering transformative solutions across neuroscience, biomedical engineering, and clinical practice. This review systematically analyzes advancements in electroencephalography-based brain–computer interface architectures, emphasizing four pillars, namely signal acquisition, paradigm design, decoding algorithms, and diverse applications. The aim is to bridge the gap between technology and application and guide future research. In signal acquisition, noninvasive systems using wet, dry, and semi-dry electrodes are more comfortable and gentler on the skin compared to traditional methods. However, ensuring stable signal quality over long periods of time remains a challenge. Minimally invasive approaches, such as microneedle arrays and endovascular probes, achieve near-invasive signal fidelity without major surgery. Paradigm design explores task-specific neural encoders. Although motor imagery paradigms are widely used in rehabilitation, they require weeks of user training. Steady-state visually evoked potential and P300 speller paradigms enable rapid calibration, but cause visual and cognitive fatigue. Advanced systems currently combine electroencephalography with electromyography or eye-tracking to better handle real-world tasks. Decoding algorithms have advanced through Riemannian geometry for improved noise filtering, deep learning architectures for automated spatiotemporal feature extraction, and transfer learning frameworks to minimize cross-subject calibration. However, challenges remain in managing inconsistent electroencephalography, reducing processing demands, and ensuring compatibility across different electroencephalography devices. Clinical trials reveal a predominant focus on stroke rehabilitation, while emerging frontiers include astronaut neuro-monitoring in space exploration. Challenges include improving signal accuracy, minimizing movement interference, addressing ethical data concerns, and ensuring real-world use. Future advancements focus on biocompatible nanomaterials, adaptive algorithms, and multimodal integration, positioning electroencephalography-based brain–computer interfaces as pivotal tools in next-generation neurotechnology.

Key words: clinical trial, deep learning, diagnosis, electroencephalography paradigms, electrode, motor imagery, rehabilitation, sensor, steady-state visually evoked potential, transfer learning