Epilepsy EEG Classification and Visualization using Linear Discriminate Analysis

Electroencephalography (EEG) is an important tool in epilepsy diagnosis practice. Many computer-based EEG analysis methods have been presented in the literature. Most of them addressed automatic classification of EEG recorded from epilepsy patients. In our previous work, a novel visualization tool for epilepsy EEG was introduced to investigate the focus based on Ensemble Empirical Mode Decomposition (EEMD) and Linear Discriminate Analysis (LDA). As well, two contours were rendered to highlight the focus of spike and seizure EEG respectively. However, it segmented EEG records into 2-second epochs and classified them independently. Some seizure epochs were falsely classified into normal. In this study, we consider EEG records continuously. The 8-second EEG in front of the current epoch is also taken into consideration. More features are extracted. Instead of using a greedy search for important features, as done in the previous work, a genetic algorithm with enforced reduction is adopted. The EEG used in this study was collected from six subjects (two normal and four with epilepsy) in National Taiwan University Hospital (NTHU). The classification accuracy is significantly enhanced. Furthermore, contours for the spike and seizure are merged into one. It is much easier for clinicians and nurses to locate the seizure focus.