Analysis of neonatal EEG signals using Stockwell transform

In this paper, we investigate the Stockwell Transform, a linear time-frequency spectral localisation technique, on non-stationary, multicomponent neonatal seizure EEG signals. The seizure signals of interest are namely slow wave and sharp spike seizures. The performance of Stockwell Transform is compared to that of existing quadratic time-frequency representation, namely the Choi-Williams Distribution and the B Distribution, on both simulated and real EEG seizure signals. The results show that the Stockwell Transform yields distinctive, interference free time-frequency patterns corresponding to the neonatal EEG seizure signals. By capturing both high-frequency spike components and predominantly low frequency components of neonatal seizures concurrently and accurately, the Stockwell Transform is able to distinguish these two types of neonatal seizure signals with unique signatures. These signatures can then be effectively used for seizure modelling, detection and prediction.