The Promise of Sleep: A Multi-Sensor Approach for Accurate Sleep Stage Detection Using the Oura Ring

Abstract

Consumer-grade sleep trackers represent a promising tool for large scale studies and health management. However, the potential and limitations of these devices remain less well quantified. Addressing this issue, we aim at providing a comprehensive analysis of the impact of accelerometer, autonomic nervous system (ANS)-mediated peripheral signals, and circadian features for sleep stage detection on a large dataset. Four hundred and forty nights from 106 individuals, for a total of 3444 h of combined polysomnography (PSG) and physiological data from a wearable ring, were acquired. Features were extracted to investigate the relative impact of different data streams on 2-stage (sleep and wake) and 4-stage classification accuracy (light NREM sleep, deep NREM sleep, REM sleep, and wake). Machine learning models were evaluated using a 5-fold cross-validation and a standardized framework for sleep stage classification assessment. Accuracy for 2-stage detection (sleep, wake) was 94% for a simple accelerometer-based model and 96% for a full model that included ANS-derived and circadian features. Accuracy for 4-stage detection was 57% for the accelerometer-based model and 79% when including ANS-derived and circadian features. Combining the compact form factor of a finger ring, multidimensional biometric sensory streams, and machine learning, high accuracy wake-sleep detection and sleep staging can be accomplished.

Keywords: accelerometer; heart rate variability; machine learning; sleep staging; wearables.

Figure 1

Technical illustration of the second generation Oura ring. The ring has a titanium cover, battery, power handling circuit, double core processor, memory, two LEDs, a photosensor, temperature sensors, 3-D accelerometer, and Bluetooth connectivity to a smartphone app.

Figure 2

Accelerometer and temperature data for one participant (Dataset 1: Singapore, 15 years old) and one night. Sleep stages annotated from PSG data are color-coded.

Figure 3

Heart rate and HRV (rMSSD) data for one participant (Dataset 1: Singapore, 15 years old) and one night. Sleep stages annotated from PSG data are color-coded.

Figure 4

Cosine, decay, and linear functions used to model sensor-independent circadian features.

Figure 5

Bland-Altman plots for total sleep time (TST), 2-stage classification, and the four models compared in this paper.

Figure 6

Epoch by epoch sensitivity for sleep and wake and the four models compared in this paper. Whiskers are computed as 1.5 times the interquartile range.

Figure 7

Bias and limits of agreement for TST, 4-stage classification, and the four models analyzed in this paper.

Figure 8

Bias and limits of agreement for time in light sleep, 4-stage classification, and the four models analyzed in this paper.

Figure 9

Bias and limits of agreement for time in deep sleep, 4-stage classification, and the four models analyzed in this paper.

Figure 10

Bias and limits of agreement for time in REM sleep, 4-stage classification, and the four models analyzed in this paper.

Figure 11

Epoch by epoch sensitivity for 4-stage classification and the four models compared in this paper. Whiskers are computed as 1.5 times the interquartile range.

Figure 12

Epoch by epoch specificity for 4-stage classification and the four models compared in this paper. Whiskers are computed as 1.5 times the interquartile range.

Figure 13

Example hypnogram for an average night (f1 = 0.78) for the model, including all features (ACC+T+HRV+C).