Multiscale spatiotemporal neural network with multi-attention mechanism using brain partitioning for motor imagery...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) aid motor-impaired individuals.
• Classifying EEG signals for motor imagery (MI) is difficult due to low signal-to-noise ratio and individual brain variability.

Purpose of the Study:

• To develop a novel neural network for improved MI classification from EEG signals.
• To enhance the integration of spatial and temporal features for more accurate brain signal interpretation.

Main Methods:

• A parallel multi-depth spatial-temporal neural network was designed, leveraging brain functional topography.
• The network utilizes two parallel branches focusing on inter-channel differences and specific brain regions (frontal, parietal).
• Specialized blocks and a multi-loss design were employed for comprehensive feature extraction and optimization.

Main Results:

• The proposed model achieved high classification accuracies: 82.14% on the BCI Competition IV 2a dataset and 95.61% on the High Gamma dataset.
• Kappa values of 0.76 and 0.93 were obtained, outperforming existing state-of-the-art methods.
• The results demonstrate superior performance in MI classification.

Conclusions:

• Parallel spatial-temporal networks utilizing brain partitioning are significant for MI classification.
• The findings support the application of this method in rehabilitation engineering and real-world BCI systems.