Energy-Efficient On-Board Processing Technique for Wireless Epileptic Seizure Detection Systems

The growth of wireless body area sensor networks (WBASNs) has led the way to advancements in healthcare applications and patient monitoring systems; epileptic seizure lies at the heart of these promising technologies. For real-time epileptic seizure detection, wireless EEG sensors have been utilized for the purpose of data acquisition, pre-processing and transmission to the server side. The dilemma of excessive power consumption of both data processing and transmission imposes strict constraints on battery-powered sensor nodes. The conventional streaming approach transmits raw EEG data as is, while consumes excessive transmission power. Other modalities consider lossy compression paradigms in order to reduce the transmitted data. This paper proposes on-board data reduction technique, which extracts low-complexity and high level, application-based, features at the sensor side. In particular, EEG spectrum is segmented to five frequency sub-bands; numerous combinations of these sub-bands are selected as feature vectors, and classification using k-nearest neighbor. Simulations have revealed that alpha and delta rhythms yield feature vectors for the EEG signals in the context of epileptic seizure detection. Satisfactory results have been obtained (around 92.47% accuracy). Moreover, the proposed approach outperforms both data streaming and compression techniques in terms of total power consumption and seizure detection performance.