[Automatic determination system of human sleep stages on an experimental basis]

Determination of the EEG sleep stages by clinicians is a time consuming task, because of a great amount of record obtained during one night. Therefore, an effective computer technique for an automatic processing EEG is strongly desired. We tried to make a real-time automatic determination system of human sleep EEG stages by a micro-computer. The system is constructed by a micro-computer which has Intel-8086 CPU, 8087 NDP and 768 KByte main memories, and A/D converter, printer, X-Y plotter and two floppy disk drives. EEG, EOG and EMG data are sampled with 50Hz sampling frequency as the input data. Stage determination is executed on basis of simplified rule of the standard one proposed by APSS, and the results are shown as the output of the printer and the X-Y plotter. The determination is executed for every one recording sheet with 20 seconds' record, by using of EEG wave pattern recognition results, REMs detection results and fluctuation of EMG level. When the result is Stage 1, final determination is deferred until next sheet in which sleep spindles or REMs are included. For EEG wave pattern recognition, autoregressive model and statistical pattern recognition method are introduced. Pattern recognition process is constructed by two parts. In one part, back-ground waves are recognized, and in the other part, sleep spindles are detected. By this system, we obtained about 70% accuracies compared with clinician's scoring. This system is real time processing system. Therefore, sleep chart is completed as the experiment is finished. This system is useful for sleep studies.