Abstract:

BACKGROUND: The sleep data are very large, and it is not satisfied with practice demand if the data cannot process by computer. However, the methods which are using at present have a disadvantage that the accuracy is comparatively low.
OBJECTIVE: To investigate a new method for sleep stage classification only using electroencephalogram (EEG) and electrooculogram (EOG) based on aperiodic waveform analysis and genetic neural network of radial basis function (RBF).
METHODS: Raw data including two-channel EEG and one-channel EOG recorded from eight subjects were obtained from Sleep-EDF database of PhysioBank, MIT. After digital filter with zero phase, raw data were analyzed by aperiodic waveform analysis to extract several parameters that were necessary for sleep stage classification. Then, preprocessed data as input for genetic neural network of RBF accepted training. Finally, test data were sent to trained neural network to validate.
RESULTS AND CONCLUSION: The results obtained, on average 95.6% of agreement between the expert and the GA-ANN for six stages of vigilance, going beyond results of known literature (70%-90%), which possess high value in practice and maybe satisfy with research and clinical application.