Real-time automated sleep scoring: validation of a microcomputer-based system for mice

Long-term circadian studies of sleep and wakefulness in rodents have been hindered by the labor required to analyze long polygraph records. To expedite such studies, we have designed and implemented SCORE, a microcomputer-based real-time sleep scoring system for rodents. The electroencephalograph is digitized in 10-s epochs at 100 Hz. Frequency and amplitude information from the waveform are extracted into a 48-dimension vector that is then compared to previously taught vectors representing the canonical features of four arousal states: wakefulness, theta-dominated wakefulness, rapid eye movement (REM) sleep, and nonREM (NREM) sleep. Match values are assigned for each state to each epoch; after excluding states based on wheel-running or drinking activity data, the nonexcluded state with the best match value for the epoch is scored. Analysis of over 23,000 epochs for four mice yielded an overall agreement of 94.0% between two human scorers and the program, compared with a 94.5% agreement between the two human scorers. The SCORE algorithm matched the human concensus best for wakefulness (97.8%) and NREM sleep (94.7%), but was lower for REM sleep (75.2%) and theta-dominated wakefulness (83.3%). Most errors in scoring of REM sleep were in close temporal proximity to human-scored REM epochs. SCORE is capable of scoring arousal states for eight animals simultaneously in real time on a standard IBM PC equipped with a commercially available analog-to-digital conversion board, and should considerably facilitate the performance of long-term studies of sleep and wakefulness in the rodent.