SleepGuard: Capturing Rich Sleep Information Using Smartwatch Sensing Data
Summary (4 min read)
1 INTRODUCTION
- Sleep plays a vital role in good health and personal well-being throughout one’s life.
- While PSG is widely considered as the gold standard for sleep monitoring, and while it is extensively used to support clinical treatments of sleep disorders, it has some disadvantages that make it unsuitable for longitudinal and large-scale sleep monitoring.
- ACM acknowledges that this contribution was authored or co-authored by an employee, contractor, or affiliate of the United States government, also known as Author’s address.
- Without details of the environment and activities across sleep stages, the root cause of poor sleep cannot be captured and the user informed of how to improve their sleep quality.
- The authors present the design and development of SLEEPGUARD (Section 2), the first holistic sleep monitoring system to rely solely on sensors available in an off-the-shelf smartwatch to capture a wide range of sleep information that characterizes overall sleep quality, user behaviours during sleep, and the sleep environment.
2 SLEEPGUARD SLEEP MONITORING PLATFORM
- SLEEPGUARD is a novel smartwatch-based sleep monitoring system that aims at estimating sleep quality and capturing rich information about behaviours and events occurring during sleep.
- By capturing this information, SLEEPGUARD can analyze potential reasons for sleep problems and provide the user with suggestions on how to improve their sleep routine or sleep environment.
- The different sensors and information extracted from them are summarized in Table 1.
2.1 Detecting Sleep Postures and Movements
- One’s sleeping position, also referred to as sleep posture, and the extent of body movements are important factors in determining overall sleep quality.
- To accomplish posture detection, the authors observe that the arm position strongly correlates to the sleep postures, i.e., the arm is typically located in a specific, stable location for a given sleep posture.
- Specifically, the authors use the first KNN model to detect if the input tilt angle reading is closest to one of their training samples where the palm was up.
- Recall that the authors are interested at detecting if the hand is placed on three positions: the head, the abdomen and the chest.
- This peak corresponds to the average respiratory frequency of an adult (0.2Hz to 0.47Hz) [1], suggesting that the PSD reading can be used as a proxy to detect respiratory events.
2.2 Labeling the Eye Movement Stages
- The authors also found that the extent of body movements can be used to judge the amplitude of respiration, which in turns allows us to detect rapid-eye-movement (REM) and non-rapid-eye-movement (NREM) stages.
- Respiratory amplitude is only an indicator of the division of the sleep stage and the authors can not regard it as a basis for final judgement, but it serves as an early reference that helps later phases of the sleep stage detection.
- In other words, a body rollover event is recorded when the posture changes are detected between two time points.
- Distinguish them from large body movements can help us to further analyze the user’s sleep stage.
- Based on this observation, the authors then use the change of accelerometer reading to distinguish between the body trembling and the hand movement.
2.3 Detecting Acoustic Events
- Acoustic events during sleep, such as snore, cough and somniloquy, can reflect and affect user’s sleep quality and physical health.
- Snore event has a continuous signal, while the cough and somniloquy are sudden events, thus the number of consecutive occurrences is very small.
- In conclusion, the “interval”, “duration” and “frequency” of acoustic events can be used as three unique features.
- The authors use the short-term average energy to calculate the peak value of the acoustic signal.
- Instead, their algorithm estimates the thresholds on a per signal basis.
2.4 Tracking Illumination Conditions
- Studies have shown that there is a significant interaction between illuminance level and the mental state of the individual [65].
- According to the study [22], it can be learned that the dim artificial light during sleep is significantly associated with the general increase in fatigue, and the proper light can be used to increase the sense of exhaustion and promote sleep.
- In other cases, when the bedroom has weak lights (e.g., when the bedroom is illuminated using a table lamp), the light sensor’s average readings are below 10 Lux.
- For most smartwatches, the light sensors are usually installed in the front face of it.
- The authors detect the wrist flip based on two aspects: (i) the rotation angle of smartwatch; (ii) whether the light intensity maintains stable after the sharp drop.
2.5 Sleep stage and quality
- Sleep is generally considered as a cyclical physiological process composed of three stages: rapid eye movement (REM) stage, light sleep stage and deep sleep stage.
- The biological characteristics of different sleep stages exhibit distinguishingly.
- The sleeper usually experiences a transition from light sleep to deep sleep and then enters REM, but sometimes there is also possible a phenomenon of skipping some certain sleep stages occurs during sleep.
- The measurement of respiratory amplitude, also known as BA(t).
3.1 Experimental Setup
- The authors evaluate SLEEPGUARD through experiments conducted in 15 single-occupancy homes over a two-week period.
- The study was approved by local IRB, and participants were separately asked to consent to release their data for analysis.
- The questionnaires were based on the Pittsburgh Sleep Quality Index (PSQI), a widely used and validated questionnaire in sleep quality research [17].
- While the performance of Fitbit is not comparable to medical grade PSG, it has been shown to have a good association in adults [14, 26, 31, 49], especially in estimating REM and light sleep stage.
- Moreover, the goal of their experiments is not to demonstrate that SLEEPGUARD is capable of medical grade sleep monitoring, but to demonstrate that it performs comparably to commercial systems in common sleep monitoring tasks, while at the same time being able to capture much richer set of sleep information.
3.2 Pilot Study: Training Data
- To train the models used in their system and to determine optimal parameter values, a small-scale pilot study with 10 participants was carried out prior to the main experiment.
- The training examples used to train their algorithms and to determine the algorithm parameters are collected from 10 users (5 males and 5 females).
- The authors testing users were asked to wear a smartwatch to sleep and collected the sensor data while they were sleeping.
- Every testing user contributes 10 nocturnal sleep data over a two-week period.
- These users were different from those took part in their evaluation (Section 3.1).
3.3 Prototype Implementation
- The smartwatch is equipped with a Quad-core Cortex-A7 processor at 1.1 GHz.
- It runs the Android Wear 2.0 operating system.
- The authors use five Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, Vol. 9, No. 4, Article 39.
- To reduce energy consumption of the smartwatch, in the experiments the authors analyse the sensor data on a a XiaoMI Note2 Android smartphone to which the smartwach sends sensor measurements over Bluetooth.
- As part of an initialization process, SLEEPGUARD estimates the initial body posture and hand position.
4.1 Evaluation of Subcomponents
- The authors focus on the detection accuracy about five events, that are body posture, the body rollover, the hand position, the micro body movement and the acoustic events.
- As the authors demonstrate in Sec. 4.2.4, these participants also suffered from poor sleep quality and hence indicate how the information extracted by SLEEPGUARD can support the detection of sleep problems.
- To assess the detection accuracy of micro body movements, the authors manually label the ground truth recorded by the camera during sleep, including hand moving, arm raising, and body trembling.
- Fig. 21 illustrates the detection accuracy across 15 users.
4.2 Overall performance
- 2.2 Effect of respiratory amplitude on sleep stage detection.
- For the remaining 6 users, the error is within one scale point, with most of the errors being between 0 and 1 (poor and general) and 2 and 3 (good and excellent).
- Events captured by SLEEPGUARD showed that this was likely due to bad hand posture as the person tended to put the hand on top of head before sleeping.
- Existing systems are only capable of capturing some of these factors influencing sleep quality and thus they are not capable of providing a holistic view of the participant’s sleep quality, whereas SLEEPGUARD is capable of providing very detailed information about sleep events.
5 DISCUSSION
- The authors have shown that sensors available on off-the-shelf smartwatches can be used to capture rich information about sleep quality and factors affecting it.
- Therefore, SLEEPGUARD is not a replacement for professional medical equipment for high-precision sleep detection, but serves as a personal technology that provides easy-to-use and non-intrusive way to monitor personal sleep patterns and to obtain feedback about the sleep quality.
- Alternatively, multi-sensor designs, such as combination of smart watch and intelligent ring, could be used to gather relevant sensor measurements from both wrists.
- Therefore, for the detection performance of sleeping posture, body rollover and hand position events has almost no effect, but it may have some influence on the body micro movements and acoustic events.
7 SUMMARY AND CONCLUSION
- The authors have presented SLEEPGUARD, the first holistic smartwatch-based sleep monitoring solution that can simultaneously estimate sleep quality and provide rich information about sleep events, including body motions, acoustic events related to sleep disorders, and ambient illumination.
- To capture this information accurately, the authors have proposed new algorithms for extracting the relevant events from sensor information.
- The authors demonstrated the benefits of SLEEPGUARD through rigorous benchmark experiments carried out using measurements collected from a two week trial with 15 participants.
- This information is particularly important for identifying possible causes of poor quality sleep and can be used to provide the user with suggestions on how to improve their sleep quality, e.g., by improving their sleep environment or behaviours surrounding sleep.
- The authors also compared the sleep quality estimates of SLEEPGUARD against subjective self assessments, demonstrating a high degree of correspondence.
Did you find this useful? Give us your feedback
Citations
51 citations
28 citations
Cites background or methods from "SleepGuard: Capturing Rich Sleep In..."
...This is because both SleepHunter and SleepGuard employ the built-in microphone of the smartwatch which may be sensitive to the location of the smartwatch and the background noise of the environment....
[...]
...It is also observed that TagSleep achieves comparable accuracy with wearable sensor-based system (SleepGuard [8])....
[...]
...The existing works [6, 17, 32, 51, 58] recognized sleep sound-activities, such as snore and cough, mainly using sound data recorded by the bulit-in microphone in the smartphone (e.g., iSleep [23], Sleep hunter [22]), smartwatch (e.g., SleepGuard [8], ubiSleep [49]) and headband (e.g., Sleep Profiler [44])....
[...]
...(iii) For accelerometer-base approaches, most sensors are built in the smartphone (e.g., Sleep Cycle [13], Sleepbot [57] and S plus [44]), smartwatch (e.g., SleepGuard [8] and Sleepmonitor [59], SleepHunter [22], FitBit [18] and Jawbone [29]) and bed pad (e.g., Hoque et al. [24] and Beddit [6])....
[...]
...For cough activity, TagSleep outperforms SleepHunter and SleepGuard with the average accuracy improvement of 30% and 10%, respectively....
[...]
23 citations
20 citations
16 citations
References
23,155 citations
"SleepGuard: Capturing Rich Sleep In..." refers methods in this paper
...The questionnaires were based on the Pittsburgh Sleep Quality Index (PSQI), a widely used and validated questionnaire in sleep quality research [16]....
[...]
21,674 citations
2,321 citations
"SleepGuard: Capturing Rich Sleep In..." refers background in this paper
...Such behaviors are most likely to occur during the deep sleep stage and the REM stage [8, 39]....
[...]
2,295 citations
"SleepGuard: Capturing Rich Sleep In..." refers background or methods in this paper
...Traditionally, the dedicated medical technologies, like EEG, ECG and EMG [67], have been applied for sleep monitoring....
[...]
...This is because polysomnography monitors and analyzes sleep based on information that directly correlates with sleep such as EEG, EMG, EOG, and oxygen saturation, whereas SleepGuard estimates sleep quality from cues that have an indirect effect on sleep quality....
[...]
...Moreover, EMG and EOG using electrodes placed on the skin near the eyes and on the muscles, respectively, measures in deeply differentiating REM stage from all the other stages....
[...]
...In PSG, medical sensors attached to human body are used to monitor events and information such as respiration, electroencephalogram (EEG), electrocardiogram (ECG), electrooculogram and oxygen saturation [29, 44, 54, 67]....
[...]
1,918 citations
Related Papers (5)
Frequently Asked Questions (14)
Q2. What future works have the authors mentioned in the paper "Sleepguard: capturing rich sleep information using smartwatch sensing data" ?
Exploring these techniques is part of their future work. This is where the authors need to measure and consider in their future work. Exploring such multi-sensor designs is an interesting future research direction. This will also be a direction for their future work.
Q3. What could be used to gather sensor data from both wrists?
multi-sensor designs, such as combination of smart watch and intelligent ring, could be used to gather relevant sensor measurements from both wrists.
Q4. What is the impact of wearing the smartwatch on a different wrist?
For the hand position detection and micro-body movement detection (including the arm raising and hand movement), wearing the smartwatch on a different wrist does have an impact.
Q5. What are the main factors that affect sleep quality?
the user’s breathing patterns, posture during sleep, and routines surrounding the bedtime also have a significant impact on sleep quality.
Q6. What is the main reason why insufficient or poor quality sleep has a significant economic burden?
Besides having an adverse effect on individuals, insufficient or poor quality sleep has a significant economic burden, among others, through decreased productivity, and medical and social costs associated with treatment of sleep disorders [47].
Q7. What are examples of consumer-grade sleep monitors?
Examples of consumer-grade sleep monitors range from apps running on smartphones or tablets to smartwatches and specialized wearable devices [10, 32, 34, 58, 67, 69].
Q8. Why is the hand placed on the head different from the other positions?
This is largely due to the upward facing direction of the palm when the hand is placed on the head compared to the downward palm direction when the hand is placed the other positions.
Q9. Why do current solutions focus on monitoring characteristics of the sleep itself?
This is because current solutions focus on monitoring characteristics of the sleep itself, without considering behaviours occurring during sleep and the environmental context affecting sleep, e.g., ambient light-level and noise.
Q10. What is the method used to detect if the hand was placed on the head?
If their hierarchical model predicts that the hand was not placed on the head, the authors then use the method described in the next paragraph to detect if it was placed on the abdomen or the chest.
Q11. What can the authors do to improve the detection performance of the sleep posture?
For this kind of situation, the authors can test the change of acceleration data in multi-sleeper situations by popularizing the experiment to adjust the detection threshold of their body’s micro movements and achieve better detection performance.
Q12. Why is SLEEPGUARD a better sleep monitoring system?
This is because PSG monitors and analyzes sleep based on information that directly correlates with sleep such as EEG, EMG, EOG, and oxygen saturation, whereas SLEEPGUARD estimates sleep quality from cues that have an indirect effect on sleep quality.
Q13. What are the main issues that need to be addressed in the system?
While the recognition performance of their system is very encouraging, there are some issues that would need to be addressed in their system before larger-scale deployment would be feasible.
Q14. What is the difference between the supine and the left-lateral posture?
In terms of errors, due to angular characteristics of acceleration being similar between the supine posture with hand putting on the head and the left-lateral posture, a small amount of the supine postures are classified as left lateral.