Actigraphy movement classification for sleep/wakefulness discrimination

Wrist actigraphy is a simple but powerful technique that allows long term monitoring of human activity. Among the wide range of areas of application, it has a special relevance in sleep studies, where its non-invasive nature makes it a valuable tool in the detection of abnormal behavioral patterns, associated with certain sleep disorders. While the detection of simple trends, such as circadian cycle phase shifts, are easily detected, an accurate detection of the sleep and wakefulness states is still an open issue. In this paper, an autoregressive model and an exponential distribution are used to model the movements recorded during nocturnal actigraphy. The parameter of the distribution, together with the coefficients of the autoregressive model and several temporal and position measures, are combined into a classification framework in order to discriminate movements during sleep and wakefulness states. It is shown that the movements recorded during each state are intrinsically different, with the fusion of classifiers achieving an accuracy of 73% in the discrimination of the two states. 1