Development of an AR-pole statistic for ECoG seizure detection

The authors consider the problem of epileptic seizure detection and focus localization from ECoG signals. ECoG are subdural recordings of electric potentials collected by invasively placing a grid on the surface of the brain. A statistic based on an autoregressive (AR) model has been developed to aid in seizure detection and localization. Specifically, the authors build a statistic based on the location of three dominant poles (obtained from the AR polynomial) relative to the unit circle. This statistic has produced results that are in agreement with findings by clinical ECoG readers for five patient cases examined.