A Spatial-Temporal Attention-Based Multi-Scale Feature Extraction Network for Motor Imagery Decoding

Zichen Ren, Peng QiIEEE Transactions on Medical Robotics and Bionics 2025 · DOI

Abstract

Brain-computer interfaces (BCIs) integrated with robotic systems has inspired some insights into various neurorobotic applications, including but not limited to exoskeleton-assisted gait, prosthesis control, robotic arm manipulation, and robotic wheelchair navigation. A pivotal limitation of this integration is its performance decoding motor imagery (MI) in electroencephalogram (EEG) signals. Existing methods for MI decoding predominantly rely on handcrafted features or ignore key spatial and temporal information, thereby extracting redundant information and eventually reducing the decoding precision. To address these issues, we propose a spatial-temporal attention-based multi-scale feature extraction network, named STANet. STANet consists of three critical modules. The multi-spatial attention module and temporal attention module enable the STANet to capture more key spatial features and temporal features, respectively. Additionally, the multi-scale feature extraction module can integrate effective features from different scales, increasing the MI decoding precision. To evaluate the effectiveness of STANet, experiments have been performed on two public datasets, namely, the BCI-IV-2a dataset and the High Gamma dataset. STANet exhibits notable classification accuracy of 80.56% and 96.87% on two datasets, outperforming the current state-of-the-art algorithms. These findings suggest that incorporating spatial-temporal attention can enhance MI decoding performance, with the potential to improve the efficiency of controlling neurorobotic applications.
BibTeX
@article{ren2025spatial,
  author = {Zichen Ren and Peng Qi},
  title = {{A Spatial-Temporal Attention-Based Multi-Scale Feature Extraction Network for Motor Imagery Decoding}},
  journal = {IEEE Transactions on Medical Robotics and Bionics},
  year = {2025},
  doi = {10.1109/tmrb.2025.3617978},
}

Loading related content...

Figure