A Spiking Neural Network With Adaptive Graph Convolution and LSTM for EEG-Based Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Electroencephalography (EEG) signal classification is crucial for brain-computer interfaces (BCIs).
• Spiking Neural Networks (SNNs) offer energy efficiency and capture complex neural dynamics for EEG analysis.
• Existing SNN methods often neglect the spatial topology of EEG channels and temporal spike dependencies.

Purpose of the Study:

• To introduce a novel SNN model, SGLNet, for improved EEG-based BCIs.
• To address limitations in capturing spatial and temporal information in current EEG analysis methods.
• To develop a more generalizable SNN model for diverse BCI tasks.

Main Methods:

• A learnable spike encoder converts raw EEG signals into spike trains.
• Multi-head adaptive graph convolution tailored for SNNs leverages spatial topology among EEG channels.
• Spike-based Long Short-Term Memory (LSTM) units are designed to capture temporal spike dependencies.

Main Results:

• SGLNet demonstrates superior performance in EEG classification compared to existing state-of-the-art algorithms.
• Empirical evaluations on emotion recognition and motor imagery decoding datasets validate the model's effectiveness.
• The model successfully integrates spatial channel information and temporal spike dynamics.

Conclusions:

• SGLNet offers a significant advancement in EEG-based BCI performance.
• The proposed model provides a new framework for developing high-performance SNNs for BCIs.
• This research highlights the potential of SNNs with graph convolution and LSTM for complex spatiotemporal EEG dynamics.