Encoding seizures with partial synchronization: A spiking neural network for biosignal monitoring on a mixed signal neuromorphic processor

Abstract

ABSTRACTLong-term monitoring of biomedical signals is crucial for modern clinical treatment of neurological disorders such as epilepsy. Encoding epileptic seizures with partial synchronization using a spiking neural network (SNN) offers a promising avenue as such networks can be implemented on ultra-low-power neuromorphic processors. Indeed, such bio-inspired neuromorphic systems containing mixed-signal asynchronous electronic circuits can perform always-on monitoring of biomedical signals for extended periods of time, without having to employ traditional clocked analog-to-digital conversion or cloud-based processing platforms.Here, we present a novel SNN architecture, co-designed and implemented with a mixed-signal neuromorphic chip, for monitoring epileptic seizures. Our hardware-aware SNN captures the phenomenon of partial synchronization within brain activity during seizures. We validate the network on a full-custom mixed-signal neuromorphic hardware using real-time analog signals converted from an Electroencephalographic (EEG) seizure data-set, and encoded as streams of events by an asynchronous delta modulation (ADM) circuit, directly integrated, together with its analog front-end (AFE) signal conditioning circuits, on the same die of the neuromorphic SNN chip.We demonstrate the ability of the hardware SNN to extract local synchronization patterns from the event streams and show that such patterns can facilitate seizure detection using a simple linear classifier. This research represents a significant advancement toward developing embedded intelligent “wear and forget” units for resource-constrained environments. These units could autonomously detect and log relevant EEG events of interest in out-of-hospital environments, offering new possibilities for patient care and management of neurological disorders.