Detection of epileptic waves in EEG based on wavelet transform.
- Author:
Chenxi SHAO
1
;
Jijun LU
;
Hao ZHOU
Author Information
1. Dept. of Computer Science, University of Science & Technology of China, Hefei 230027.
- Publication Type:Journal Article
- MeSH:
Electroencephalography;
Epilepsy;
diagnosis;
Humans;
Wavelet Analysis
- From:
Journal of Biomedical Engineering
2002;19(2):259-272
- CountryChina
- Language:Chinese
-
Abstract:
Detection of epileptic waves in EEG is particularly helpful in the interpretation of the underlying process in seizures. This study is aimed at providing a new method for automatic detection of epileptic waves through the wavelet analysis of EEGs. It mainly deals with the detection of spikes or spike-waves based on wavelet transform. Since spikes and spike-waves contain high frequency energy, they will be represented in a particular scale localized in a small time window. According to these feature waveforms of epileptic waves, a continuous processing system for epileptic waveforms detection is constructed. We apply discrete wavelet transform on EEGs. Because of the time-frequency domain localization of wavelet transforms, we can get the local maximal positions across several successive dyadic scales of wavelet transform. And these positions indicate the points of sharp transitions in EEGs. Then we calculate the distance between every two successive maximal positions in each scale. This distance stands for the period of subwave. Furthermore, the distribution of subwave periods of each scale can be worked out. Then, comparing the distribution of normal EEG's and epileptic EEG's. The difference between these two waveforms provides us the criteria for automatic detection and classification. In order to reduce the detection workload, we also compare the detection efficiency of each scale. The scale that provides highest accuracy is selected for our automatic detection system. The results presented in this study show that scale 3 provides the best detection accuracy. So, scale 3 is deemed to be the proper scale for automatic detection. This system has the following advantages: (1) Reduced the workload significantly by selecting proper scale(s) for automatic selection; (2) Enhanced the detection accuracy by selecting proper criteria and threshold; (3) Capable of continuous detection; (4) It is also fit for the detection of other biomedical signals. This system showed good performance, and the initial clinical results obtained are also encouraging.
