Segmentation of Electroencephalographic Signals using an Optimal Orthogonal Linear Prediction Algorithm

Abstract

Finding the stationary parts of an EEG signal proves to be a difficult task because of the various shapes of the signal that can be found for different subjects, both normal and abnormal. The paper presents the results obtained in the segmentation of an EEG signal into stationary parts in order to be used in the stages of feature extraction and classification of these signals or in predicting epileptic seizures. For this purpose an optimal orthogonal linear prediction algorithm that implements a so-called “innovations filter” is used. The algorithm was tested on EEG signals from epileptic patients and the results are presented in detail. Some further applications and possible research fields are also offered. Ill. 8, bibl. 7 (in English; summaries in English, Russian and Lithuanian).