Seizure prediction using cross-correlation and classification

Prediction of seizures is a difficult problem as the EEG patterns are not wide-sense stationary and change from seizure to seizure, electrode to electrode, and from patient to patient. This paper presents a novel patient-specific algorithm for prediction of seizures in epileptic patients. Cross-correlation coefficients are extracted every 2 seconds using a 4-second window with 50% overlap from focus electrodes identified by the epileptologist. Features are further processed by a second-order Kalman filter and then input to three different classifiers which include AdaBoost, radial basis function kernel support vector machine (RBF-SVM) and artificial neural network (ANN). The algorithm is tested on the long-term intra-cranial EEG (iEEG) database collected at the UMN epilepsy clinic. This database includes EEG recordings from 2 patients sampled from varying number of electrodes sampled at 2kHz. It is shown that the proposed algorithm achieves a high sensitivity and a low false positive rate.