EEG Feature Extraction Accelerator Enabling Long Term Epilepsy Monitoring Based on Ultra Low Power WSNs

Abstract

Wireless sensor networks based on miniaturized ultra low power nodes are increasingly utilized in highly demanding cyberphysical systems required to offer extended time periods of non-intrusive and unattended monitoring of specific biophysical signals. However, although significant performance enhancements are offered by state of the art WSN platforms, still scarce energy availability comprises the Achilles’ Hill of respective platforms. A prominent approach to remedy this deficiency is to transfer part of the overall processing applied on raw data into the WSN node, aiming to reduce drastically the amount of data than need to be wirelessly transmitted, thus allowing the radio chip to be turned off most of the time. The proposed approach has been applied in the context of ARMOR project, which focuses on EEG based Epilepsy monitoring and automated seizure detection, considering that seizure detection is based upon specific EEG feature vectors representing just ~5 % of the total amount of raw data. In that respect, this paper presents the design, implementation and evaluation of a novel EEG feature vector extraction hardware accelerator module. The validity of the module’s functionality is verified against already published Matlab based seizure detection algorithms while performance benefits are also estimated.