Skip to Main Content
Traditionally the analysis of sleep has used two distinct manual EEG analysis methods: one for general structure, the other for short time-scale events. Both methods suffer from high levels of inter-expert variability. In this paper we present a system which uses a neural network classifier to analyse each second of sleep. Postprocessing techniques are described which result in outputs which mimic both of the traditional manual analysis methods. This combination of methods results in a comprehensive sleep analysis system providing information on both the macro and microstructure of sleep. Our results show that it is possible to use a combined approach to sleep analysis and that there is strong correlation between expert scoring and the post-processed neural network output.