An attention-based motor imagery brain–computer interface system for lower limb exoskeletons.

This article focuses on the development and evaluation of an attention-based brain–computer interface (BCI) system designed to control a customized lower limb exoskeleton for rehabilitation purposes. The proposed system integrates convolutional neural networks (CNN) with a lightweight attention module to decode motor imagery electroencephalogram (MI-EEG) signals corresponding to left- and right-leg movements. Experiments involving eight participants demonstrated that this method outperforms several existing CNN-based approaches in offline classification accuracy and achieves an average online control accuracy of approximately 60%, validating the system’s feasibility for practical use. The study highlights challenges such as low signal-to-noise ratio and subtle differences between left- and right-leg MI-EEG signals, and suggests that the exoskeleton’s physical feedback may enhance user engagement and decoding performance.