Statistical features-based comparison of analysis and synthesis of normal and epileptic electroencephalograms for various wavelets

An electroencephalogram (EEG) is an electrical signal in microvolts captured noninvasively from the brain, which provides important and unique information about the brain. The frequency of an EEG signal lies between 0 and 100 Hz. Decomposition of an EEG signal into various bands such as alpha, beta, delta, theta, and gamma is essential in seizure-related studies. EEGs play a key role in the diagnosis of epileptic seizures and neurological disorders. In this paper, multiple wavelet families for decomposition and reconstruction are explored and are compared based on risk functions and reconstruction measures. While dealing with the wavelets it is a difficult task to choose the correct/accurate wavelet for the given biosignal analysis. Various statistical properties were studied by the authors to check the suitability of various wavelets for normal and diseased EEG signal decomposition and reconstruction. The methodology was applied to 3 groups (63 subjects) consisting of both sexes and aged between 1 and 80 years: 1) normal healthy subjects, 2) patients with focal seizures, and 3) patients with generalized seizures. Our result shows that the Haar and Bior3.7 wavelets are more suitable for normal as well as diseased EEG signals, as the mean square error, mean approximate error, and percent root mean square difference of these wavelets are much smaller than in other wavelets. The signal-to-error ratio for Haar and Bior3.7 was much higher than in any other wavelet studied.