Predicting EEG seizures using graded spiking neural networks.

Area of Science:

• Neuromorphic Computing
• Computational Neuroscience
• Epilepsy Research

Background:

• Epileptic seizures pose significant challenges to patient quality of life.
• Existing seizure prediction systems often struggle with real-time processing and energy efficiency.
• Development of non-patient-specific, efficient prediction models is crucial for widespread clinical application.

Purpose of the Study:

• To develop and evaluate a novel, non-patient-specific epileptic seizure prediction system.
• To implement this system using graded spiking neural networks (GSNNs) on Intel's Loihi 2 neuromorphic processor.
• To address challenges of real-time, energy-efficient seizure prediction.

Main Methods:

• Utilized the CHB-MIT dataset for training GSNNs.
• Integrated hyperparameter optimization and EEG channel selection for data reduction.
• Employed a multi-windowed voting mechanism for enhanced robustness.
• Deployed the system on Intel's Loihi 2 neuromorphic processor.

Main Results:

• Achieved a non-patient-specific prediction accuracy of 99.14%.
• Reached a throughput of 21.6 EEG segment inputs per second with 25.104 mJ energy consumption per input.
• Demonstrated significant improvements in event and synaptic communication sparsity compared to ANNs.

Conclusions:

• Introduced a robust and energy-efficient GSNN-based framework for epileptic seizure prediction.
• Significantly enhanced potential for real-time, wearable seizure prediction applications.
• Highlighted the promise of GSNNs in advancing neuromorphic computing for epilepsy management.