Realisation d'un systeme automatique permettant de coder le sommeil des nourrissons (2 A 12 MOIS)

Abstract

We have developed a system which permits automatic analysis of sleep of infants from 2 to 12 months of age. It is based on the calculation of a score on the basis of 6 parameters weighted according to their importance in determining the different phases of sleep. We use 3 EEG parameters derived from spectral analysis of the C zO z derivation: total spectral energy (Hjorth activity), energy of the delta band, and energy of the beta 1 band, and 3 polygraphic parameters: variability of muscle activity in each minute, variability of the respiratory frequency and presence or absence of rapid eye movements. The analysis provides a list of the sleep phases, minute by minute, and a traditional sleep diagram; we also calculate the successive appearances of the different sleep phases, in 10 min periods, and schematize these results in a diagram which shows the general trend of each sleep phase during the night. When compared with visual analysis, these results are satisfactory, since we observe on average 86.2% agreement in the differentiation of calm, paradoxical and transitional sleep, and 76.4% agreement in the differentiation of stages II, III and IV of calm sleep.