Real-time sub-milliwatt epilepsy detection implemented on a spiking neural network edge inference processor.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) signal analysis for epileptic seizure detection faces technological challenges in timely diagnosis.
• Existing methods struggle with real-time processing and efficient power consumption for portable applications.

Purpose of the Study:

• To develop and validate a spiking neural network (SNN) for detecting interictal and ictal periods of epileptic seizures.
• To evaluate the performance of the SNN on the Xylo neuromorphic processor for low-power, real-time seizure detection.

Main Methods:

• Utilized a trained spiking neural network (SNN) to analyze electroencephalogram (EEG) datasets.
• Deployed the SNN model on the Xylo neuromorphic processor, simulating spiking leaky integrate-and-fire neurons.
• Compared the SNN's performance against existing studies and evaluated its accuracy and power consumption.

Main Results:

• The SNN achieved high test accuracies of 93.3% for ictal and 92.9% for interictal period classification.
• The deployed system demonstrated low average power consumption: 87.4 μW (IO power) + 287.9 μW (compute power).
• The method exhibited excellent low-latency performance across multiple datasets.

Conclusions:

• The proposed SNN approach offers an effective solution for online, real-time epileptic seizure detection.
• The low-power and low-latency characteristics make it suitable for integration into portable and wearable diagnostic devices.
• This technology has the potential to significantly improve the management and monitoring of epilepsy.