Publication: Seizure detectetion in eeg signals using stationary wavelet transform and multi layer perception neural network
Loading...
Date
2012-06-01
Authors
Ho, Qiao Yee
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In this study, an algorithm is presented for the analysis and classification of the Electroencephalogram (EEG). EEG is a signal record of the brain electrical activity, which has been widely used in the investigation of neuropathology, such as the epileptic seizure. The epileptic seizure is a sudden and temporary electrical disturbance of the brain causing involuntary movements, muscular contractions, convulsions, loss of consciousness, etc. However, the abnormalities in the non-stationary EEG are times too delicate to be detected and analyzed using conventional methods, such as the Fourier Transform (FT). This study has described the use the Stationary Wavelet Transform (SWT) in analyzing the epileptic characteristics of the EEG signal. Through the scheme of multiresolution analysis (MRA), the transient features of the EEG signal can be accurately captured and localized in both the temporal and spectral contexts. The proposed EEG signal processing is divided into two pre- and post- parts; both are worked on the basic of wavelet transform. In the EEG pre-processing, the proposed
Heuristic SureShrink Denoising (HSD) and Intuitive Kurtosis Despiking (IKD) techniques are applied to remove the noises and artifacts that reside in the EEG signal, whereas the EEG post-processing is essentially the EEG features extraction. In the context of EEG classification, an Artificial Intelligence (AI) classifier is applied, which is the feedforward Multi Layer Perceptron Neural Network (MLPNN). The extracted features are used as the inputs to the MLPNN, and the error-backpropagation (BP) training algorithm is applied to train the MLPNN to classify three types of EEG: normal, interictal and ictal. The classification performance of the network is evaluated in terms of the sensitivity, specificity and accuracy by using the 3-Channel Confusion Matrix. The obtained results have shown a classification accuracy of 97.22%. This figure has confirmed that the proposed algorithm has potential in classifying the EEG signals.