Automatic sleep analysis: myth or reality?

The increase in the number of sleep recordings has gone hand in hand with the development of sleep analysis methods designed to automatically generate a hypnogram. These automated analyses yield satisfactory results in healthy subjects, but remain inadequate for a large number of subjects recorded in sleep centres. The aim of this book is to clarify the reasons for these shortcomings and to demonstrate the value of spectral and non-linear analyses in the assessment of human sleep. We demonstrate the contribution of spectral analysis to the overall assessment of the organisation of the night and of the activation dimension (AD) - a parameter derived from the correlation dimension - in the more detailed analysis of sleep stages. The DA values show that sleep stages are heterogeneous depending on the brain regions recorded and the age of the subjects. These data suggest that numerical analyses of the sleep EEG should be utilised more effectively than simply constructing a hypnogram. Doctors, scientists and engineers involved in the study of sleep should find in this book a comprehensive overview of the issues surrounding sleep analysis.