Activity-based sleep–wake identification in infants

Abstract

Actigraphy offers one of the best-known alternatives to polysomnography for sleep-wake identification. The advantages of actigraphy include high accuracy, simplicity of use and low intrusiveness. These features allow the use of actigraphy for determining sleep-wake states in such highly sensitive groups as infants. This study utilizes a motion sensor (accelerometer) for a dual purpose: to determine an infant's position in the crib and to identify sleep-wake states. The accelerometer was positioned over the sacral region on the infant's diaper, unlike commonly used attachment to an ankle. Opposed to broadly used discriminant analysis, this study utilized logistic regression and neural networks as predictors. The accuracy of predicted sleep-wake states was established in comparison to the sleep-wake states recorded by technicians in a polysomnograph study. Both statistical and neural predictors of this study provide an accuracy of approximately 77-92% which is comparable to similar studies achieving prediction rates of 85-95%, thus validating the suggested methodology. The results support the use of body motion as a simple and reliable method for determining sleep-wake states in infants. Nonlinear mapping capabilities of the neural network benefit the accuracy of sleep-wake state identification. Utilization of the accelerometer for the dual purpose allows us to minimize intrusiveness of home infant monitors.

Algorithms for sleep–wake identification using actigraphy: a comparative study and new results

“…PS is the predicted probability of sleep. Sazonov et al (2004) reported the following performance with their algorithm: 75.4% accuracy, 93.2% sensitivity, 41% specificity.…”

“…In this section, we will describe how one can apply two previously reported actigraphy algorithms (Sadeh et al, 1994;Sazonov et al, 2004) to the CHIME data set. These two algorithms use linear combinations of features computed from the actigraphic signal to distinguish sleep from wake.…”

“…These two algorithms use linear combinations of features computed from the actigraphic signal to distinguish sleep from wake. As the original algorithms in Sadeh et al (1994) and Sazonov et al (2004) were optimized for different actimeters than the one used in our chosen data set, we will recalculate the parameters in these algorithms using the CHIME data set before testing them for their predictive capabilities. The results of applying these algorithms to our chosen data set are presented in the Results section.…”

“…Notable among the reported actigraphy algorithms are the ones developed by Sadeh et al (1989Sadeh et al ( , 1994Sadeh et al ( , 1995 covering not only adult normal and abnormal subjects, but also infants, the algorithm of Cole et al (1992), the Actigraph Data Analysis Software (adas) by Jean-Louis et al (1996), the algorithm reported by Sazonov et al (2004) and various commercially available algorithms accompanying the different actimeters available in the market.…”

“…This paper is organized as follows. The following section introduces the Collaborative Home Infant Monitoring Evaluation (CHIME) database, the several quality indicators that will be used to assess the performance of the actigraphy scoring algorithms, and the algorithms of Sadeh et al (1994) and Sazonov et al (2004). Next, the proposed pattern recognition methods, the multilayer perceptron (MLP) and the decision tree are introduced in a tutorial fashion.…”