Actigraphic Detection of REM Sleep Based on Respiratory Rate Estimation

Abstract— The use of wrist actigraphy in sleep research has for long been limited to the classification of sleep/wake; little progress has been made in the evaluation of the sleep stages. We propose and evaluate two novel algorithms: a method for respiratory rate estimation based on spectral analysis of actigraphic data, and a method for estimating REM sleep based on the detected respiratory rates. Using simultaneous recordings of polysomnography and actigraphy data acquired from 34 subjects, we found that our proposed method successfully estimated respiratory rate with low mean absolute error (0.52 counts/min), and REM sleep with high positive predictive value (64.5%), but low sensitivity (11.0%). While the low sensitivity hinders the immediate clinical use of our algorithms, our findings are important in indicating for the first time that actigraphs have the potential to detect REM sleep.