Use of kNN and quadratic discriminant analysis methods for sleep staging from single lead ECG recordings

Abstract

Sleep consists of REM and four non-REM stages. Determining a person's sleep stage in a certain part of night sleep is performed by the technical experts using the polysomnographic recordings acquired in special sleep laboratories. The acquisition of these recordings for the sleep characterization require not only the connection of various sensors and electrodes to the subject but also spending the night in a bed which is different from the subject's own bed. In this study we investigated the feasibility of using only an electrocardiographic holter device instead of a polysomnography system used in a sleep laboratory for the sleep study and phase determination. For this purpose, single lead ECG data obtained during the night sleep (mean sleep duration 7 hours) from 18 subjects (6 men) with ages between 20 and 67 were used for sleep staging based on R-R interval values. The validation was performed by the sleep stage data previously determined by the sleep experts. Phase determination consists of R-R interval computation, feature extraction and classification studies. The features used in this study were the median value, the difference between the 75 and 25 percentile values, and mean absolute deviations of the R-R intervals computed in each 30-second epoch. The k nearest neighbor (kNN) and quadratic discriminant analysis methods based on one-versus-others approach were used as the classification tools. In the testing procedure cross-validation was employed. As a result, out of awake stage and other five sleep stages four stages were classified accurately at a rate of greater than 80%.