Contactless Sleep Apnea Detection on Smartphones
Summary (5 min read)
Contributions:
- The authors make four key contributions: (1) The authors introduce a novel contactless technique for tracking chest and abdomen movements due to breathing on smartphones.
- The authors achieve this by analyzing the reflections from FMCW sonar transmissions.
- (2) We design algorithms to detect central apnea, obstructive apnea, and hypopnea as well as estimate the apnea-hypopnea index from the sonar reflections.the authors.the authors.
- (3) We implement their design on off-the-shelf smartphones and demonstrate the ability to concurrently track breathing movements from multiple subjects.the authors.the authors.
- (4) We perform a clinical study with 37 patients demonstrating the feasibility of using their system to accurately compute the number of central, obstructive, and hypopnea events as well as the apnea-hypopnea index.the authors.the authors.
2. POLYSOMNOGRAPHY OVERVIEW
- The clinical polysomnography test (PSG) is traditionally used to diagnose sleep apnea and other sleep disorders.
- Physicians classify the sleep apnea level using these AHI values.
- The scoring process of analyzing these epochs involves two main steps.
- Fig. 3 (b) plots the nasal pressure and chest motion signals during a hypopnea event.
- Throughout a sleep duration of eight hours, the technician monitors the sensors and ensures that they remain properly attached to the patient's body.
3. APNEAAPP
- ApneaApp is a contactless system that enables detection of sleep apnea events using smartphones.
- To understand how ApneaApp operates, the authors first describe how they transform the phone into an active sonar system that tracks the chest and abdomen movements due to breathing.
- The authors then describe their algorithms to detect sleep apnea events from these movements.
3.1 Transforming the Phone into an Active Sonar
- These signals reflect off the reflector (e.g., human body) and arrive at the microphone after a time delay.
- The chirp duration, Tsweep, in practice is picked so that the reflections from all points within the desired operational distance would start arriving before the chirp ends.
- To search for these breathing movements in each FFT bin, the authors perform another FFT over a 30s duration and search for peaks in the typical breathing frequencies of 0.2-0.3 Hz.
- Note that the above procedure is repeated every time the user moves their position, which the authors identify using the algorithm in §3.2.2.
3.2 Sleep Apnea Detection Algorithm
- As described in §2, diagnosing sleep apnea requires estimating the Apnea-Hypopnea Index (AHI) which is the average rate of apnea events during the sleep duration.
- This requires computing the number of central, obstructive, and hypopneas as well the total sleep time.
- The authors first describe their algorithms to compute the number of apneas and then the total sleep time.
3.2.1 Estimating the Number of Apneas
- ApneaApp detects a hypopnea event when the chest motion reduces below a threshold (30%).
- But to identify central apnea, the authors couple an amplitude reduction in the chest motion signal with an absence of the breathing periodicity.
- The authors run the peak detection algorithm to identify the locations of the peaks in the chest motion signal.
- The authors again use the peak detection algorithm to detect the peaks.
- To compute this threshold for their sonar data, the authors perform a linear regression on the data from a single patient to max- imize the hypopnea detection accuracy and identify a threshold of 38%.
3.2.2 Estimating the total sleep time
- In a polysomnography test, the EEG sensors are used to measure the brain activity to determine whether the subject is asleep or awake; this is then used to measure the total sleep time of a subject.
- Recent work [30] has shown that one can use body movements to compute the total sleep time and the resulting accuracies are acceptable for the purposes of estimating the apnea-hypopnea index.
- To identify these movements, the authors leverage that when the subject's body moves, the sonar reflections experience large variations that do not have the periodicity of the breathing motion.
- The authors then check if the resulting peaks are aperiodic.
- For periodic breathing motion, this derivative is close to zero but is higher for aperiodic signals.
4. IMPLEMENTATION
- The authors implement ApneaApp as a third party Android app that does not require rooting the smartphone.
- 1 shows that all the tested smartphones satisfy this requirement.
- This happens because of a feedback loop between the co-located microphone and speaker and can result in sleep apnea misclassifications.
- To address this issue, the authors leverage that many smartphones today come with an additional microphone that is not co-located with the speaker.
- This means that smartphones that do not have this additional microphone, such as the Galaxy Nexus, cannot be used to accurately detect sleep apnea events with ApneaApp.
5. CLINICAL STUDY
- The authors conducted a clinical sleep study with 37 patients (17 female and 20 male) between ages of 23-93 (mean: 50) for a total of 296 hours and compare the results from ApneaApp with the PSG study reports.
- The sleep lab is equipped to conduct PSG studies for a maximum of eleven patients per night; the authors randomly choose up to five subjects per night for their study.
- The authors do not screen patients based on their gender, race or national origin, but only consider adults.
- The patients who participated in their study were not provided any monetary benefits.
- Further, there was significant interest amongst them in using a smartphone instead of the existing PSG procedure to diagnose their sleep illness.
Protocols.
- In clinical PSG studies, the sleep lab technician assigns each patient to a separate room furnished with an adjustable king size bed.
- Chin electromyogram and right and left electroculogram electrodes were also applied.
- The position and the orientation of the phone vary across patients depending on their sleep habits.
- The data from the clinical PSG study is sent to a third-party entity that scores the sensor data and provides the number of central and obstructive apneas, and hypopneas.
- Thus, the authors compare it with ApneaApp's signals using Matlab.
5.1 Sleep Apnea Detection Accuracy
- The key metric used for sleep apnea diagnosis is the Apnea-Hypopnea Index (AHI), which represents the average rate at which apnea and hypopnea events occur during the sleep duration.
- First the authors examine ApneaApp's effectiveness in computing the number of central apnea, hypopnea, and obstructive apnea events.
- The authors use the body motion sensors from the PSG study as their ground truth.
- Next, the authors evaluate how well this correlates with the sleep time computed in the PSG study using the brain activity (EEG sensors).
- The five misclassifications occur between no-apnea and mild-apnea; four of them happen right at the boundary between the two levels with an error less than 1 event/hr.
6. MICROBENCHMARKS
- The authors run experiments with five participants (ages 24-25) in a home bedroom environment to evaluate the effects of various parameters such as orientation, position and distance.
- The bedroom is next to a major street with significant foot and vehicular traffic.
- Since the authors cannot imitate the sleep apnea events where the chest motion experiences amplitude variations, in these set of experiments they use the coarse-grained breathing frequency as a proxy to understand the effects of the various parameters.
- To obtain the ground truth data for the breathing frequency, the authors use a Vernier respiratory belt worn at the abdomen level.
6.1 Effect of Phone Distance, Orientation and Position
- The authors first evaluate the effect of the phone distance, orientation and position on the breathing frequency accuracies.
- This is because, as described in §4, the breathing frequency on the phone is relatively stable compared to the amplitude variations that are necessary for detecting sleep apnea events.
- It also limits the negative effects of environmental changes farther away than a meter on ApneaApp's accuracies.
- Introducing human conversations in the vicinity of the experiments also does not affect these accuracies.
- Next, the authors experiment with the phone at different positions around the subject.
6.2 Effect of Sleeping Position and Blankets
- Next, the authors evaluate the accuracies for different sleeping positions and in the presence of blankets.
- Fig. 14 shows that the average residual error is below 0.16 breaths per minute across all the sleeping positions.
- The authors note that the accuracy is lower when the patient is lying with her face down .
- The authors measure the breathing frequency accuracy for various blanket thicknesses.
6.3 Breathing Signals from Multiple Subjects
- As discussed in §3.1, the sonar reflections from multiple subject arrive at different times at the microphone.
- Thus, ApneaApp can simultaneously track breathing movements from more than one sub- ject.
- To evaluate this, the authors monitor the breathing rate from two subjects sleeping in the supine position on the bed.
- Fig. 16 plots the accuracies for the two subjects as a function of the distance between them.
- This is expected, because as described in §3.1, their FMCW transmissions can be used to resolve reflections with a greater resolution than 20 cm.
6.4 Evaluating ApneaApp's Audibility
- ApneaApp transmits FMCW audio signals in the 18-20 kHz range.
- To evaluate the audibility of these signals, the authors ran ApneaApp with 144 subjects: 37 of whom were sleep apnea patients who took part in their clinical study, 50 additional sleep apnea patients not part of their clinical study, and 57 healthy undergraduate subjects who are seniors in the UW CSE department.
- None of the 87 sleep patients reported any audible sounds from the phone.
- Further, the 37 patients who took part in their clinical study had no audibility issues for the eight-hour sleep duration.
- One of the undergraduates could hear faint sounds when the phone was placed next to his ear and a second undergraduate reported hearing a sound similar to TV static at distances of up to one meter.
8. LIMITATIONS AND FUTURE DIRECTIONS
- In addition to diagnosing sleep apnea, a PSG test is also used to detect other events such as the respiratory-effort related-arousals .
- A direction worth exploring is to leverage their sonar design with other smartphone sensors to detect these non-sleep apnea conditions.
- The authors note however that in principle their sonar design can also be used to extract the heart rate without requiring contact with the human body.
9. CONCLUSION
- The authors present a contactless solution that detects sleep apnea events on smartphones.
- To achieve this, the authors introduce a novel technique that can track chest and abdomen movements due to breathing on smartphones using FMCW sonar transmissions.
- The authors design algorithms to detect central apnea, obstructive apnea, and hypopnea as well as estimate the total sleep time by analyzing the FMCW reflections from the human body.
- The authors implement their design on off-theshelf smartphones and demonstrate the ability to concurrently track breathing movements from multiple subjects.
- Finally, the authors present results from a clinical study with 37 patients that demonstrate the feasibility of using their system to accurately compute the number of central, obstructive, and hypopnea events as well as the apneahypopnea index.
Did you find this useful? Give us your feedback
Citations
347 citations
Cites background or methods from "Contactless Sleep Apnea Detection o..."
...Finally, ApneaApp [27] tracks the periodic breathing movements in a sleep environment using FMCW reflections of the inaudible transmissions from the phone....
[...]
...To appreciate the challenge, microphones on today’s mobile devices have a sampling rates of 48 kHz [27]....
[...]
318 citations
Cites background from "Contactless Sleep Apnea Detection o..."
...Another system, ApenaApp, uses chirp signals to detect the changes in reflected sound that is caused by human breaths [18]....
[...]
252 citations
Cites background from "Contactless Sleep Apnea Detection o..."
...In [22] authors analyze sounds emitted while the user is sleeping to signal sleep apnea episodes....
[...]
228 citations
Additional excerpts
...[25] leverages acoustic FMCW to detect chest and abdomen movements, which is used to identify sleep apnea....
[...]
218 citations
Cites background from "Contactless Sleep Apnea Detection o..."
...Moreover, nearly one-third (30%) of Canadian adults between 18 and 79 years of age were estimated to be at intermediate or high risk for sleep apnea [126], which is often associated with high blood pressure, heart failure, diabetes, stroke, attention deficit/hyperactivity disorder, and increased automobile accidents [127,128]....
[...]
...[127] Contactless Sleep Apnea Detection Samsung Galaxy S4 Phone speaker and micro-phone • The speaker transmits 18–20 kHz sound waves and the microphone senses the reflections....
[...]
References
9,642 citations
"Contactless Sleep Apnea Detection o..." refers background in this paper
...Building on the above system, we develop algorithms that identify various sleep apnea events including obstructive apnea, central apnea, and hypopnea from the sonar reflections....
[...]
1,263 citations
"Contactless Sleep Apnea Detection o..." refers background in this paper
...…in computing of rate of apnea and hypopnea events is as low as 1.9 events/hr. CATEGORIES AND SUBJECT DESCRIPTORS J.3 [Computer Applications]: Life and Medical Sciences GENERAL TERMS Design; Human Factors; Algorithms KEYWORDS Mobile Health; Sleep Apnea; Phone Sonar; Contactless Breathing Monitoring...
[...]
573 citations
"Contactless Sleep Apnea Detection o..." refers background in this paper
...…in computing of rate of apnea and hypopnea events is as low as 1.9 events/hr. CATEGORIES AND SUBJECT DESCRIPTORS J.3 [Computer Applications]: Life and Medical Sciences GENERAL TERMS Design; Human Factors; Algorithms KEYWORDS Mobile Health; Sleep Apnea; Phone Sonar; Contactless Breathing Monitoring...
[...]
571 citations