Classification of Seizures in EEG Using Wavelet-Chaos Methodology and Genetic Algorithm

• Main Authors: Kai-Cheng Hsu, 許凱程
• Other Authors: Sung-Nien Yu
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/90926245286433215382

碩士 === 國立中正大學 === 電機工程所 === 97 === Detection of seizures in EEG can be challenging because of myogenic artifacts and might be time-consuming when reviewing long term EEG recordings. In this study, we propose a method based on wavelet-chaos methodology and genetic algorithm for automatic seizure detection in EEG. The data used in this research are obtained from both healthy and epileptic subjects and are available online from the University of Bonn, Germany. In the experiment, the wavelet transform was used to decompose EEG into five EEG subbands that approximate to delta, theta, alpha, beta, and gamma subbands. Non-linear parameters, including the time lag, the embedding dimension, the correlation dimension, and the largest Lyapunov exponent, were extracted from each of the frequency band and the original EEG signals. The nonlinear parameters were employed as the features to train the Bayesian classifier (BC), the support vector machine with linear kernel function (SVML) and the support vector machine with radial basis function kernel function (SVMRBF) classifier. Finally, a genetic algorithm (GA) was used for selecting the effective feature subset to improve the performance of the classifiers. Three groups of EEG recordings, including the EEG from healthy subjects (group A), the interictal (group B) and the ictal (group C) EEGs from epileptic subjects, were recruited for the study. The EEG signals were classified into two groups, i.e. seizure-free (group A+B) and seizure (C) groups, and three groups, i.e. A, B, and C, separately. According to the experimental results, the average accuracies of the BC, the SVML, and the SVMRBF with GA are 89.99%, 90.17%, and 90.38%, respectively. The GA is also used for parameters optimization and weights adjustment for sensitivity and specificity. The results demonstrate that the classifiers with nonlinear features are effective in seizure classification in EEG. The performance of classifiers is improved when the GA-based optimal feature subset selection method is employed.