A method for automatic REM sleep segmentation

A basic property of REM sleep is that it is not a uniform stage. Two kinds of variations are particularly significant. First, a remarkable amount of literature, based on polygraphic signals, fMRI images, and psychophysiological data, has shown that “phasic” epochs, presenting distinct oculomotor activity, should be distinguished from “tonic” epochs. Second, recent literature has provided interesting data about slow eye movements (SEMs), which are present during REM sleep in addition to rapid eye movements (REMs), although, according to Rechtschaffen & Kales, during REM sleep SEMs “do not approach the abundance, magnitude, and pendular pattern of the slow eye movements during stage 1”. Our research has aimed to propose a method for segmenting REM sleep into three substages, respectively characterized by (a) enhancement of REMs; (b) selective enhancement of SEMs; (c) no enhancement of either REMs or SEMs. A wireless digital polygraph (SOMNOScreen) was used for recording polygraphic signals from healthy subjects while they were sleeping in their houses. Twelve EEG traces were recorded from electrodes placed over the scalp according to the 10-20 System, together with two EOG traces, EMG, and EKG. Our method consisted of the following stages: (a) band-pass filtering in order to extract a slower component (0.2-0-6 Hz) and a faster (1-3 Hz) component from the original EOG signal; (b) recognition of events consisting in a transient increase of either of these components; (c) calculation of parameters characterizing these events; (d) insertion of these parameters into a relational database; (e) final segmentation provided by queries to this database. This method allowed quantitative parameters to be calculated connected to the amplitude variations of REMs and SEMs during each REM sleep epoch in the course of the night. The study of the oscillating properties of eye movements can lead to a better knowledge of the thalamocortical intrinsic loop active during REM sleep and of the interaction between unimodal sensorimotor areas and higher-order association cortices. Furthermore, an investigation of the properties of the REM substages in patients affected by REM sleep Behavior Disorder can shed light on the mechanisms of this disease, thus helping clinicians with the diagnosis and the treatment. Other currently debated issues connected with the study of REMs and SEMs regard vulnerability during REM sleep, the homeostatic process of sleep regulation, and the features of mentation in the course of sleep.