Automatic real-time analysis of human sleep stages by an interval histogram method.

A new interval histogram method for automatic, all-night sleep stage scoring, simulated on a digital computer, is described. The system consists of a 2-step analysis. The first step is recognition of elementary patterns in EEG, EOG and EMG, and the second step is the determination of sleep stages based on these parameters. Correlation of this method with power spectral analysis of the dominant EEG patterns during each sleep stage supported the reliability of the first step analysis. Overall agreement (89.1%) between the computer and human judges was only 3% less than the agreement (92.1%) among the scorers, indicating considerable reliability of the second step. The primary areas of disagreement that arose in the identification of sleep stages occurred with stages 1, 2 and REM. To improve scoring accuracy, the system may require epoch sequence information. The profile of the elementary parameters of the EEG signals clearly illustrated the cyclic nature of these activities throughout the night. The alpha and delta 2 waves clearly separated the awake state from sleep stages. Beta 2 can discriminate stages 1 and REM from stage 2, and the best indicator for distincting stage 1 from REM was muscle activity. Sigma and spindles were prominent during stage 2 sleep. Both delta 2 and high voltage delta waves distinguished stage 3 from stage 4. On the other hand, delta 1 was evenly distributed and seemed to be common to all sleep stages.