Epileptic Seizure Prediction Based on Synchroextracting Transform and Sparse Representation
Epilepsy is a serious brain disorder, affecting more than 50 million people worldwide. If epileptic seizures could be predicted in advance, patients can take measures to avoid the unfortunate consequences. However, seizure prediction still remains an unsolved problem. One important approach of seizure prediction research is feature extraction and classification using electroencephalography (EEG) signals. This research focuses on classifying epileptic EEG signals efficiently so that an epileptic seizure can be predicted accurately prior to the episode. According to our experiments, permutation entropy is found to be very sensitive to the epileptic stages, and it can be used for screening and selecting electrode channels on a scalp from which EEG signals are obtained. Sample test predicted all of six seizures at least one hour before the episodes in one patient (ID 1) EEG in this study. In addition, synchroextracting transformation (SET) and mutual information (MI) are employed to improve the time-frequency resolution as high as possible. SET is a more energy concentrated time-frequency representation than classical time-frequency analysis methods. Sparse representation (SR) is also applied for the accurate EEG signal stages classification. SR is a parsimonprinciple that a sample can be approximated by a sparse linear combination of basis vectors. The sample tests reached the accuracy of 99% and 98% in seizure stage classification based on SR combined with SET transform.