Localized Comparison of Sleep Stage Scoring Between PSG and Wearable Devices

Abstract

Wearable devices have been recently proposed as an alternative to the gold standard—polysomnography system (PSG)—in several particular circumstances. The past decade has seen a renewed importance in this area, resulting in various researches attempted to evaluate the validity and reliability of this kind of surrogate devices.​ This paper seeks to address the issue of consensus in results reached by both wearable devices and PSG system. Specifically, we propose an algorithm to locally detect and quantify the discrepancies among the wearable devices and PSG system in scoring sleep stages. Within this study, the developed validation algorithm was performed using the data collected from 14 human subjects, ​each of whom wears 4 wearable devices simultaneously to retrieve the signals of interest associated​ with PSG system as a benchmark overnight at our sleep lab ​for sleep monitoring and record. The gist of the  algorithm is to compute the differences between two consecutive points in the hypnogram—a graph that represents the stages of sleep over the time—to estimate the monotone correlation of the sleep stages between wearable device and PSG. Our main goals are to determine sleep periods where the most considerable discriminants happen, and to quantify the levels of disagreement between two sorts of the instrument. Preliminary results show that the localized correlation algorithm can detect the points of the most significant agreement and disagreement of wearable devices benchmarked to PSG. The validation method for this algorithm will soon be an inevitably critical issue to be verified and developed for the robustness and reliability of wearable devices in comparison with gold standard PSG system.