Machine classification of infant sleep state using cardiorespiratory measures.

We examined the potential to classify sleep and waking state over the first 6 months of life in normal infants using only cardiac and respiratory measures. Twelve hour all-night polygraph recordings which included EEG, eye movement, whole body movement, facial muscle electromyographic, cardiac, and respiratory activity from 25 normal infants were collected at 1 week, and at 1, 2, 3, 4, and 6 months of age. Each minute of these recordings was classified into quiet sleep, waking, or rapid eye movement sleep by trained observers using EEG and somatic criteria. Respiratory rate and variability, heart rate and variability, and cardiac interbeat interval variation at respiratory and lower frequencies from 12 of the 25 infants were used as measures in discriminant analyses of sleep state for test on the 13 remaining infants. Using all 7 cardiac and respiratory measures, sleep states were classified with an accuracy approximating that attained by trained observers who had available all polygraph tracings (84.8% overall correct classification). Using only cardiac measures, the accuracy of classification decreased slightly, with an overall correct classification of 82.0%. Using only respiratory measures, the accuracy of classification diminished further, with an overall correct classification of 80.0%. Cardiac and respiratory measures provide quantifiable indications of sleep and waking states over the first 6 months of life in normal infants.