Local binary pattern based feature extraction and machine learning for epileptic seizure prediction and detection

Abstract

Electroencephalogram (EEG) signals have been extensively utilized to identify brain disorders such as epilepsy. Local descriptor based feature extraction methods have been utilized to generate features from EEG signals in order to understand them. The best known local descriptor is the local binary pattern (LBP). The LBP was presented for images and, at the same time, the one-dimensional form was presented for signals to use the advantages of the LBP. The advantages of the LBP are as follows: (i) it has a low computational complexity, (ii) it generates valuable features and (iii) it is easily implemented. In order to benefit from these advantages, many LBP-like local descriptors have been presented for both images and signals. In this chapter we discuss the effect of one-dimensional LBP and LBP-like descriptors on EEG signal recognition. We also explain the multilevel feature extraction model using LBP-like descriptors for EEG signals. In this chapter an LBP based feature extraction framework for epileptic seizure prediction and detection is presented. Feature selection/reduction is the crucial stage of EEG signal classification. Neighborhood component analysis (NCA) and ReliefF based feature reduction methods are explained. Finally, different machine learning techniques will be compared for epileptic seizure prediction and detection.