计算机科学
解码方法
脑电图
人工智能
语音识别
神经解码
模式识别(心理学)
脑-机接口
计算机视觉
编码(内存)
作者
Yi-Feng Chen,Mengjiao Zhang,Mingjie Dong,Zhenghong Li,Yuzhou Lin,Yusheng Li,Changsong Zhou,Zhiqiang Zhang,Jian S. Dai,Mingming Zhang
标识
DOI:10.1016/j.eng.2026.10.015
摘要
Variations in latency between trials disrupt temporal alignment in electroencephalography (EEG) and reduce the reliability of motor decoding. Much of this variation originates in the timing of endogenous neural processes; global alignment methods cannot adequately address it when they assume that all components undergo the same latency shift. We propose a component-wise temporal alignment framework that treats EEG as a combination of functionally distinct temporal components, each with its own latency profile. The framework estimates the latency of individual components, realigns them separately, and recombines them to reconstruct temporally coherent event-related potentials (ERPs) rather than shifting entire trials rigidly. In unimanual and bimanual motor tasks, this procedure markedly improved the temporal organization of EEG responses: Positive ERP peaks increased ∼6-fold and negative peaks ∼3-fold, while phase consistency across trials almost doubled. Evaluation with eight decoding architectures showed that the alignment-based strategies attained 100.6% and 97.6% of the optimized sliding-window baseline accuracy for the unimanual and bimanual tasks, respectively. These findings show that component-level correction of latency variability recovers temporally coherent neural dynamics and supports near-optimal decoding without computationally demanding exhaustive temporal-window searches, providing a general framework for EEG-based motor brain–computer interfaces.
科研通智能强力驱动
Strongly Powered by AbleSci AI