A novel morphology-based classifier for automatic detection of epileptic seizures

Most of the automatic seizure detection schemes reported in the literature are complex for detecting seizures that are of (a) short duration, (b) minimal amplitude evolution, or (c) non-rhythmic mixed frequency epileptic activity. We present a novel morphology-based classifier to detect epileptic seizures for intracranial EEG recording. The method characterizes epileptic seizure by detecting continual presence of sharp half-waves in the EEG. Performance is evaluated on single channel intracranial EEG of seven patients, and compared to two previously developed methods for intracranial EEG recordings by our research group. The method detects seizure of varying types (rhythmic, non-rhythmic, short- and long- seizures) with a sensitivity of 100%, a false detection rate of 0.1/h and an average onset delay of 9.1 s. The method outperforms the two previously developed methods and is computationally simple for real-time application. Preliminary results on seven patients data are very promising.