The Study Of Nonlinear EEG Feature Extraction Methods

Epilepsy is one of the most common neurological disorder, which is char-acterized by recurrent and transient epileptic seizures. It typically manifests in muscle stiffness, staring and impaired consciousness etc., and may increase the risk of sustaining physical injuries and even result in death. The accurate seizure detection has been considered to be the most important step for the diagnosis and treatment follow-up in epilepsy patients. However, the traditional visual examination of long-term EEGs by a trained neurologist is a time consuming and subjective process. Therefore, there has been an increasing interest in the study of the automated seizure detection using EEGs in recent years. In order to realize it successfully, how to design an appropriate feature extraction method is the important topic in the study. This paper proposes three novel feature ex-traction methods, including lagged-Poincare-based feature extraction method LPBF, fusion feature extraction method MS-SE-FF and fuzzy conditional Renyi entropy-based feature extraction method FCRE. Then combining extreme learn-ing machine (ELM), three automatic seizure detection methods have been built respectively. The main results are presented as follows:Chapter 1 systematically introduces the background and development of automatic seizure detection methods.Chapter 2 presents some most common used feature extraction methods and classifiers in the study of the automatic detection of epileptic seizures.Chapter 3 first proposes a novel lagged-Poincare-based feature extraction method LPBF by characterizing the lag-T Poincare plots of EEGs from the quantitative point of view. The experimental verification is then shown.Chapter 4 first designs a new Mahalanobis-similarity-based feature extrac- tion method MS-F. Then a novel fusion feature MS-SE-FF by incorporating MS-F and sample entropy (SE-F) is proposed to further improve the performance. The experimental verification are finally shown.Chapter 5 first proposes a new fuzzy conditional Renyi entropy-based fea-ture extraction method FCRE on the basis of fuzzy conditional Renyi entropy. The experimental verification is then shown.