Online Epileptic Seizure Detection in Long-term iEEG Recordings Using Mixed-signal Neuromorphic Circuits

Seizure detection stands as a critical aspect of epilepsy management, which requires continuous monitoring to improve patient care. However, existing monitoring systems face challenges in providing reliable, long-term, portable solutions due to the computational expense and power demands of continuous processing and data transmission. Edge computing offers a viable solution by enabling efficient processing locally, close to the sensors and without having to transmit the sensory signals to remote computing platforms. In this work, we present a mixed-signal hardware implementation of a biologically realistic Spiking Neural Network (SNN) for always-on monitoring with on-line seizure detection. We validated the hardware system with wideband intracranial Electroencephalography (EEG) signal recordings with over 122 continuous hours of data, without pre-filtering. The network was tested with a cohort of 5 patients and a total number of 22 seizures including generalized and focal onsets. Our system effectively captures spatiotemporal features based on synchronized multichannel intracranial EEG activity, achieving 100% sensitivity across all patients and near zero false alarms. Remarkably, inference across patients required only calibrating the parameters of the network’s output layer on a single recorded seizure from each patient.