Newborn EEG Seizure Simulation Using Time-Frequency Signal Synthesis
Nathan J. Stevenson, L. Rankine, Mostefa Mesbah, B. Boashash · 2005
This paper presents a new method of simulating electroencephalogram (EEG) signals induced by a particular form of newborn seizure. The technique utilises time– frequency signal synthesis. The simulation is based on a nonstationary multicomponent waveform with piecewise linear frequency modulation (LFM). The time–dependent spectral magnitude of the piecewise LFM multicomponent signal is assigned a slowly oscillating envelope and used to construct a time–frequency image. The time–frequency image is used to synthesise a time-domain signal using the modified short–time Fourier transform (MSTFT) magnitude method. The simulated seizures are varied according to several parameters outlined in the literature to provide a large database of EEG seizures. A comparison of the spectrograms of simulated and real seizure results in an average, two–dimensional correlation coefficient of 0.8 (N=5).