Reference:
D Wilson, DD Rowlands, DA James, T Cutmore, Nonlinear Signal Processing of Electroencephalograms for Automated Sleep Monitoring, Proceedings of SPIE 5651, 95 Abstract: An automated classification technique is desirable to identify the different stages of sleep. In this paper a technique for differentiating the characteristics of each sleep phase has been developed. This is an ideal pre-processor stage for classifying systems such as neural networks. A wavelet based continuous Morlet transform was developed to analyse the EEG signal in both the time and frequency domain. Test results using two 100 epoch EEG test data sets from pre-recorded EEG data are presented. Key rhythms in the EEG signal were identified and classified using the continuous wavelet transform. The wavelet results indicated each sleep phase contained different rhythms and artefacts (noise from muscle movement in the EEG); providing proof that an EEG can be classified accordingly. The coefficients founded by the wavelet transform have been emphasised by statistical techniques. Hypothesis testing was used to highlight major differences between adjacent sleep stages. Various signal processing methods such as power spectrum density and the discrete wavelet transform have been used to emphasise particular characteristics in an EEG. By implementing signal processing methods on an EEG data set specific rules for each sleep stage have been developed suitable for a neural network classification solution. |
Publications > 2005 >