Abstract
An automatic sleep scoring system was developed by recording sleep EEG and EOG from 265 subjects. The automatic method is based on the use of two channel electro-oculogram (EOG).  Synchronous EEG activity in S2, S3 and S4 was detected by calculating the cross correlation between two EOG channels and the peak to peak amplitude in the 0.5-6 Hz band using two different thresholds. Band 1.5-6 Hz and automatic detection of fast and slow eye movements (SEM) was used to separate wakefulness, S1 and REM. Beta power 18-30 Hz was used for artefact detection. Data of 133 subjects was used to determine the optimum detection thresholds and validation of the scoring system was performed with the data of 132 different subjects. By using simple detection and smoothing rules the Cohen's kappa agreement to the standard visual scoring system was moderate (0.50) and the epoch by epoch agreement was 64 % in separating wake, S1, REM, S2, S3 and S4 sleep stages.  The advantage of the limited bandwidth automatic method is that it could be applied during online recordings using only three disposable self-adhesive electrodes. Further improvements in classification performance are possible e.g. by spindle analysis and more advanced scoring rules which make better use of temporal information.