Seizure detection using wavelet decomposition of the prediction error signal from a single channel of intra-cranial EEG

This paper presents a novel patient-specific algorithm for detection of seizures in epileptic patients from a single-channel intra-cranial electroencephaolograph (iEEG) recording. Instead of extracting features from the EEG signal, first the EEG signal is filtered by a prediction error filter (PEF) to compute a prediction error signal. A two-level wavelet decomposition of the prediction error signal leads to two detail signals and one approximate signal. Eight features are extracted every one second using a 2-second window with a 50% overlap. These features are input to two different types of classifiers: a linear support vector machine (SVM) classifier and an AdaBoost classifier. The algorithm is tested using the intra-cranial EEG (iEEG) from the Freiburg database. It is shown that the proposed algorithm can achieve a sensitivity of 95.0% and an average false positive rate (FPR) of 0.124 per hour, using the linear SVM classifier. The AdaBoost classifier achieves a sensitivity of 98.75% and an average FPR of 0.075 per hour. These results are obtained with leave-one-out cross-validation. In addition, for 13 out of 18 patients, the AdaBoost classifier requires only one feature, while it requires 4 features for the remaining 5 patients.