Nonlinear models for detecting epileptic spikes

We present a technique for automatic detection of epileptic spikes in electroencephalogram (EEG) recordings. We use a nonlinear modeling method based on information theory that enables us to defect rapidly and accurately epileptic behavior in the EEG signal. An optimal embedding dimension of the model is determined by the minimum in the mean square error between EEG signals and the corresponding model prediction. Our approach is illustrated by an application to two EEG time series: (i) interictal activity from a focal epileptic patient, and (ii) a petit mal from a generalized epilepsy patient. [S1063-651X(98)15612-X].