Spatio-Temporal Attention with Spiking Neural Networks for Seizure Detection from Electroencephalogram Signals

Area of Science:

• Computational Neuroscience
• Medical Signal Processing
• Artificial Intelligence in Medicine

Background:

• Automated seizure detection from Electroencephalogram (EEG) signals is crucial for epilepsy management.
• Current deep learning methods, while accurate, suffer from high energy consumption, limiting clinical application.
• There is a need for energy-efficient yet accurate seizure detection algorithms.

Purpose of the Study:

• To introduce a novel Spatio Temporal Attention with Spiking Neural Networks (STASNN) architecture.
• To leverage the energy efficiency of Spiking Neural Networks (SNNs) combined with attention mechanisms for seizure detection.
• To reduce the computational and energy demands of automated seizure detection systems.

Main Methods:

• Developed STASNN, integrating Spatial Attention with SNN (SASNN) and Temporal Attention with SNN (TASNN) modules.
• SASNN captures inter-channel spatial dependencies in EEG data.
• TASNN captures long-range temporal dynamics, encoding information via sparse binary spikes for energy efficiency.

Main Results:

• STASNN demonstrated significantly reduced theoretical energy consumption compared to traditional ANNs.
• The proposed method achieved high seizure detection accuracy on benchmark datasets (CHB-MIT and Siena).
• STASNN outperformed existing state-of-the-art seizure detection methods in performance and efficiency.

Conclusions:

• STASNN offers a promising solution for energy-efficient and accurate automated seizure detection.
• The integration of attention mechanisms with SNNs provides a powerful framework for biomedical signal processing.
• The developed architecture has the potential for practical clinical deployment in epilepsy monitoring.