ABSTRACT
Electroencephalographic (EEG) has become a successful means of seizure detection in adults. This usually involves identifying sharp repetitive waveforms that indicate the onset of seizure. In adults, these EEG seizures are easily recognizable against a low amplitude random background characteristic of normal brain activity. The initial patterns obtained by a time–frequency (TF) analysis of EEG seizure signals specifically confirm that EEG seizures in newborn are well characterized by a linear frequency-modulated (FM) or piecewise linear FMs. The characterization of nonstationary EEG signals in the TF domain is the first step toward an automatic method of seizure detection and classification that use powerful tools of TF signal processing. The EEG data collected thus far show that neonatal EEG seizures are highly nonstationary and occasionally multicomponent, and are mostly concentrated in the band of frequency. Seizure detection using the autocorrelation method relies on the assumption that the essential characteristic in newborn seizure EEG is periodicity.
