Tuning method of modified wavelet transform in human sleep EEG analysis

We have developed an automatic discrimination system of human sleep EEG (electroencephalogram) stages based on a wave-shape recognition method. These systems can detect discrete stages (stage MT, W, 1, 2, 3, 4 and REM). But, it is impossible to extract much information in detail by them. Therefore, we tried to continuous wavelet analysis applied to EEG signals in order to extract more precise information for the stages. We have proposed a modified wavelet transform method and an extraction method of time series of peak frequency based on time-frequency analysis. It was confirmed that the modified wavelet transform method was effective, although it was necessary to tune the parameter in the modified wavelet transform in order to get more effective features. In this paper, tuning method of modified wavelet transform is discussed from viewpoint of the feature extraction about human sleep EEG stages. Two types (linear and nonlinear) of tuning methods are proposed. And it is confirmed that nonlinear tuning method based on sigmoid function is effective through the experimental studies.