Seizures are often the first sign of neurological disease or dysfunction in the human newborn. Their clinical manifestation however, is often subtle, which tends to hinder their diagnosis. This represents an undesireable situation since the failure to quickly and accurately diagnose seizure can lead to long term brain injury or even death. In this paper, the problem of automatic seizure detection in the newborn based on the electroencephalogram (EEG) is considered. It is shown that good detection performance of electrographic seizure, which is the manifestation of seizure within the EEG, is possible using a new approach which is based on a model for the generation of the EEG. This model is derived from the histology and biophysics of a localized portion of the brain and is thus physically motivated. The model for EEG seizure is first presented along with an estimator for the model parameters. Then a seizure detection scheme based on the model parameter estimates is suggested. The method is then used to detect seizure in both simulated and real newborn EEG data. This approach gives superior performance over conventional classification approaches which rely on training data to produce observable test statistics. This is because, in general, trained classifiers are particularly susceptible to the extreme variability of the EEG over time as well as from patient to patient.
展开▼
