Seizure detection in newborns’ EEG signals using time-scale analysis

Abstract

The discrete wavelet transform (DWT) has proven one of the most suitable tools for analyzing non-stationary signals such as EEGs. In this paper, a novel automated method for detecting seizures in the newborns using the DWT of the EEG signals is proposed. The detection scheme is based on observing the changing behavior of the statistical quantities of the wavelet coefficients (WCs) of the EEGs at various scales. The used statistical quantities of the EEGs are: the number of zero-crossings, the average distance between zero-crossings of the adjacent WCs, the number of extrema, and the average distance between extrema of the adjacent WCs of certain scales. These statistics form a feature set of size 8 for each EEG segment of 6 seconds length. The extracted feature set is then fed to an artificial neural network (ANN) classifier to organize the EEG signals into seizure and nonseizure activities. The study population consists of 6 neonates’ EEGs, with ages ranging from 2 days to 2 weeks. The obtained results show the high performance of the selected feature set by the average seizure detection rate of 95%.