Wearable Epilepsy Seizure Detection on FPGA With Spiking Neural Networks.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy monitoring requires balancing accuracy, unobtrusiveness, and real-time performance.
• Electroencephalography (EEG) signals are time-varying, necessitating advanced modeling techniques.
• Spiking Neural Networks (SNNs) show promise for modeling brain states from EEG data.

Purpose of the Study:

• To develop an extremely lightweight SNN-based solution for real-time epilepsy seizure detection.
• To achieve high detection accuracy and efficiency comparable to state-of-the-art methods.
• To assess the model's suitability for deployment on wearable monitoring devices.

Main Methods:

• Utilized a simple encoding scheme to create a sparse and lightweight SNN.
• Evaluated the SNN model on the CHB-MIT EEG dataset for seizure detection.
• Assessed real-time inference performance on the SYNtzulu platform for wearable device suitability.

Main Results:

• Achieved 96% Area Under the Curve (AUC) and 99.3% average accuracy.
• Detected 100% of seizure events with a low false alarm rate of 0.3 per hour.
• Demonstrated ultra-low inference time (0.5 μs) and energy consumption (4.55 nJ) for wearable applications.

Conclusions:

• The proposed lightweight SNN offers a highly accurate and efficient solution for epilepsy seizure detection.
• The model's performance and low resource requirements make it ideal for real-time monitoring on wearable devices.
• This approach advances the development of practical, everyday epilepsy monitoring solutions.