A Nonlinear Model of Newborn EEG with Nonstationary Inputs

Abstract

Newborn EEG is a complex multiple channel signal that displays nonstationary and nonlinear characteristics. Recent studies have focussed on characterizing the manifestation of seizure on the EEG for the purpose of automated seizure detection. This paper describes a novel model of newborn EEG that can be used to improve seizure detection algorithms. The new model is based on a nonlinear dynamic system; the Duffing oscillator. The Duffing oscillator is driven by a nonstationary impulse train to simulate newborn EEG seizure and white Gaussian noise to simulate newborn EEG background. The use of a nonlinear dynamic system reduces the number of parameters required in the model and produces more realistic, life-like EEG compared with existing models. This model was shown to account for 54% of the linear variation in the time domain, for seizure, and 85% of the linear variation in the frequency domain, for background. This constitutes an improvement in combined performance of 6%, with a reduction from 48 to 4 model parameters, compared to an optimized implementation of the best performing existing model.