Seizure prediction using EEG segmentation change points

Researches show that changes in the normal activity of nervous system develop minuets to hours before the actual onset of epileptic seizures due to abnormal neural discharges. In this paper, the quasi-stationary nature of Electroencephalogram (EEG) signal and coupling between different signal channels is studied with the purpose of predicting epileptic seizure. A high rate of coinciding change points is observed before seizure onset between different electrodes after dividing signal into quasi-stationary segments. Averaged prediction time obtained varies between 10-40 minutes for 10 subjects of "CHB-MIT Scalp EEG" database.