Hand rehabilitation exoskeleton system based on EEG spatiotemporal characteristics

Abstract

Hand motor impairments can severely impact stroke survivors’ activities of daily living (ADL). To assist in restoring hand function, this paper designed and developed a hand rehabilitation exoskeleton system based on the motor imagery (MI) paradigm. To improve the accuracy of MI classification, we integrated EEGNet with graph convolutional networks (GCNs) to extract spatiotemporal and topological features. A four-layer GCN was used to enhance classification accuracy. Accurate and rapid recognition of MI enables efficient active and passive hand rehabilitation training. Additionally, we used brain network analysis by examining changes in the connection weights of topological edges over time. This approach enhances the model’s interpretability, facilitating better observation of the rehabilitation process and providing valuable insights for healthcare professionals. Offline experimental results demonstrate that, compared to traditional methods, our proposed electroencephalography (EEG) signal recognition approach significantly improves classification accuracy, achieving 89.84$\text{\%}$ ± 2.80$\text{\%}$; in the online training experiment, the average completion rate was 87.75$\text{\%}$ ± 3.88$\text{\%}$. This system effectively helps patients with hand motor impairments in completing hand rehabilitation training.