Modelling of ECoG in temporal lobe epilepsy.

Subdural recordings of the electrical activity of the human brain give electrocorticograms (ECoG) almost free of artifacts and distortions by the skull and other intervening material. This paper discusses the modelling of the ECoG during the pre-ictal, ictal and post-ictal phases of an epileptic seizure. Optimum order linear autoregressive (AR) models are formed and the movement of the poles of the models are traced with time. Nonlinear extension to the AR models (NAR) is formulated based on the assumption of existence of nonlinear oscillations in the data. The optimum order of this model is determined and its performance is compared with that of the linear AR models. The analysis of the data with NAR resulted in the satisfaction of sufficient conditions for limit cycles in the ictal phase.