This paper presents and clinically validates a new method to accurately classify sleep phases within a wrist-worn device (e.g., smartwatch, ?tnessband). The method combines inertial and optical sensors to compute the wearer’s motion, breathing rate, and pulse rate variability, and to estimate the different sleep stages (WAKE, REM and NREM). The presented method achieves a sensitivity and speci?city for the REM of 89.2 % and 77.9 % respectively; for the NREM class 83.4 % and 84.9 % respectively; and a median accuracy of 81.4 %. The assessment of the performance was obtained by comparing to the gold standard measure in sleep monitoring, polysomnography.
EMBEC & NBC 2017, H. Eskola, O. Väisänen, J. Viik, J. Hyttinen (Eds.), Singapore (Singapore), pp. 615-618