Techniques previously designed for electroencephalographic (EEG) seizure detection in the newborn have been relatively inefficient due to their incorrect assumption of local stationarity of the EEG. To overcome the problem raised by the proven nonstationarity of the EEG signal, current methods are extended to a time-frequency (TF) approach [8, 10]. This allows the analysis and characterization of the different newborn EEG patterns, the first step toward an automatic TF seizure detection and classification. An in-depth analysis of the previously proposed autocorrelation and spectrum seizure detection techniques identified the detection criteria that can be readily extended to the TF domain. We present the various patterns of observed TF seizure signals and relate them to current specialist knowledge of seizures. In particular, initial results indicate that a quasilinear instantaneous frequency (IF) can be used as a critical feature of the EEG seizure characteristics. These findings led to propose a TF-based seizure detector. This detector performs a two-dimensional (2D) correction between the EEG signal and a reference template selected as a model of the EEG seizure in TF domain.
|Title of host publication||Applications in Time-Frequency Signal Processing|
|Number of pages||32|
|Publication status||Published - Jan 1 2002|
ASJC Scopus subject areas