COMPUTER CLASSIFICATION OF SLEEP IN PRETERM AND FULL-TERM NEONATES ATSIMILAR POSTCONCEPTIONAL TERM AGES

A classification strategy of neonatal sleep is being developed by comp aring visually scored minutes of 21 channels of electroencephalographi c (EEG)/polygraphic recordings with the corresponding values for each physiological signal derived from either visual or computer analyses. Continuous 3-hour sleep studies on 54 preterm and full-term neonates a t similar postconceptional term ages were acquired under environmental ly controlled conditions using a computerized monitoring system. An on -line event marker program recorded behavioral observations. One of th ree EEG sleep states was assigned to each of 8,995 minutes by traditio nal visual analysis criteria. EEG spectral values, spectral and nonspe ctral cardiorespiratory calculations and behaviorally observed movemen ts, arousals and rapid eye movement counts were submitted for discrimi nant analysis. Based on the total minutes known for each of three stat es (i.e. active, quiet and awake), linear combinations of all specifie d digitized parameters were formed into an arithmetic algorithm by use of discriminant analysis, which served as the basis of a state assign ment for each minute. Fifty percent of the data were arbitrarily used as the training set to derive the state classification model. The rema ining fifty percent of the data were used as the cross-validation ''te st sample'' to determine the accuracy of the classification when compa red to the visually analyzed score for each corresponding minute. Thir teen out of 32 physiological measures best predicted state for both pr eterm and full-term neonatal groups. For both groups, the correct clas sification for active sleep was 90.3%, quiet sleep was 97.4%, awake wa s 97% and the overall accuracy was 93.3%. However, the order of signif icance for specific variables differed between these two neonatal grou ps. Differences in the order of variables that predict sleep states be tween preterm and full-term infants may reflect adaptation of brain fu nction of the preterm infant to prematurity and/or prolonged extrauter ine experience.