Parametric Models for Characterization, Quantification and Defection of Epileptform Events in the Electroencephalogram

Abstract

This work presents an automated method based on the autoregressive (AR) modelling of the electroencephalogram (EEG). The EEG signal is divided in short segments (typically 2 seconds) and AR models subsequently evaluated. The model parameters quantify each segment and constitute features for classification using pattern recognition techniques. The method was validated with a data set including three types of epilepyic signals: petit mal (3 hours and 45 minutes; 7 patient), interictal spikes (10 minutes; 1 patient) and partial complex seizures (2 hours; 2 patients). (...)