Neuromorphic Computing Driven Wearable Seizure Prediction System Using Spiking Neural Networks and Biosignals

Abstract

Epileptic seizures are rather unexpected events that significantly affect the quality of life of the patient which certainly should be equipped with stable systems that will give early signals. Conventional seizure predictive frameworks are very expensive to compute and is limited to real time operation. This study suggested developing an energy-efficient and wearable seizure-monitoring device by taking advantage of neuromorphic computing and spiking neural network (SNN) to process real-time biosignals (e.g., EEG, ECG). Raw bio signals are filtered and processed into spike trains and fed to a neuromorphic SNN, whereby features and time patterns of the biosignals are calculated. Benchmark data have been assessed using the proposed system and the rate of accuracy of the prediction and respectively the sensitivity with 90.1 and the specificity with 93.2 are achieved more than doing so using conventional deep learning models as well as low power usage capabilities that would be applicable to a wearable deployment. These results demonstrate the possibility of integrating neuromorphic architectures with biosignal monitoring to predict seizures in a personalized and real-time mode which is an interesting approach to improving the safety and autonomy of patients.