An EEG-based Brain Computer Interface For Automatic Sleep Stage Classification And Its Applications

Facing with the growing pressure of life and work,more and more people are forced to join in the fast-paced lifestyle.And meanwhile,more people suffer from sleep disorders,the quality of sleep heads into a sharp decline.Under such background,developing the study of sleep activities and improving the quality of sleep are of great significance.Sleep staging is an important part in the field of sleep research.Through sleep staging,the process of individual sleep activity becomes clear.The duration of slow wave sleep,the total duration of sleep,the efficiency of sleep and other evaluation indicators of sleep quality can be calculated easily.Sleep staging is an effective way to help the assessment of sleep quality.At present,the sleep staging based on EEG becomes the dominant trend in the field of sleep staging.As for the analysis and processing of the EEG signal,the Brain-computer Interface(BCI)is a typical representation.In this study,we have combined the BCI with the sleep staging innovatively,and proposed an EEG-based Brain-computer Interface for automatic sleep stage classification.(1)Based on the analysis of frequency domain and statistical calculation,we proposed an automatic sleep staging algorithm.Specifically,we first extract the spectrum characteristics and the spectral edge frequency of the EEG signal with the usage of Fourier transformation.And then,calculate the mean value,the standard deviation and other statistical characteristics.We concatenate the features mentioned above to obtain the feature vector.To validate the performance of our algorithm,we applied it on the public databases using a cross validation scheme,and then compare with related works.The average classification accuracy is 85%,and the Kappa coefficient is 0.83.The experimental results demonstrated the efficiency of our algorithms.(2)Design the off-line experiment of sleep staging.Record the EEG signal and the EOG signal according to the AASM.The EOG signal is used for assisting the sleep specialist to finish the label evaluation and the EEG signal is used for further analysis.Based on the proposed algorithm,the feature vector was extracted and then the cross validation was applied to test the generalization performance of the proposed algorithm.Using the cross-validation scheme,we have achieved an average accuracy of 77%,and the Kappa coefficient is 0.7.(3)Design the on-line experiment of sleep staging.Based on the BCI approach to record the sleep signals of the subject in a real-time way,and then presented the real time prediction with the usage of pre-trained model.By combining the on-line results predicted by system with the true labels scored by experts,we can calculate the confusion matrix and on-line accuracy,and then evaluate the performance of the on-line sleep staging system.The online average accuracy is 77%with the Kappa coefficient of 0.68,and the system performance was thus demonstrated.