Improving Actigraph Sleep/Wake Classification with Cardio-Respiratory Signals

Actigraphy for long-term sleep/wake monitoring fails to correctly classify situations where the subject displays low activity, but is awake. In this paper we propose a new algorithm which uses both accelerometer and cardio-respiratory signals to overcome this restriction. Acceleration, electrocardiogram and respiratory effort were measured with an integrated wearable recording system worn on the chest by three healthy male subjects during normal daily activities. For signal processing a Fast Fourier Transformation and as classifier a feed-forward Artificial Neural Network was used. The best classifier achieved an accuracy of 96.14%, a sensitivity of 94.65% and a specificity of 98.19%. The algorithm is suitable for integration into a wearable device for long-term home monitoring.