VALIDATION OF COMPUTER ANALYZED POLYGRAPHIC PATTERNS DURING DROWSINESS AND SLEEP ONSET

A computer system for the automatic analysis of polygraphic records wa s validated. Records from 9 subjects made during routine MSLT tests we re analysed both by two preliminary and one consensus scorer and by a computer system. Special attention was paid to the analysis of drowsin ess periods. Therefore a classification system including 7 stages, thr ee for wakefulness and movement, one for drowsiness and three for the sleep stages Sl, S2 and SREM was used. Adaptive segmentation was used to divide the records into short segments of variable length (mean 1.6 sec, range 0.5-13.7 sec). The agreements between the computer and vis ual scores were relatively good for 5 subjects having a prominent occi pital alpha activity during wakefulness (range 70-79%) but less promis ing (range 64-70%) for the other 4 subjects with ''poor'' occipital al pha activity. The values obtained corresponded to the inter-scorer agr eements. Most of the discrepancies were between adjacent stages. At ti mes in the presence of strongly fluctuating EEG amplitudes and especia lly with the ''low-alpha'' subjects it was very difficult to determine exactly even by visual scoring when, for instance, drowsiness became sleep. It is concluded that the reliability of the system is sufficien t for practical purposes especially if critical parts of the records a re visually reexamined. It was found to be difficult to define unambig uous scoring criteria for subjects with poorly defined EEG rhythms giv ing insufficient landmarks for stage determination.