Rapid eye movement detection in infants using a neural network

Counting of rapid eye movements (REM) during sleep represents one of the criterions for sleep stage scoring. Though numerous investigations have been carried out there is a lack of reliable procedures to replace the manual evaluation of sleep stages. The authors present a new and robust algorithm using a neural network based approach. It is suitable for the daily clinical use in a childrens' hospital sleep laboratory. An adaptive signal preprocessing distinguishes between REM induced signals and artefacts. The supervised training has so far been verified using polysomnographic recordings of 16 infants. EOG based determination of sleep stages are in good correspondence with EEG data and the course of the heart rate variability. The new algorithm will be part of the authors' polysomnographic diagnostic system POLDI.