Top Banner

Click here to load reader

iSleep: Unobtrusive Sleep Quality Monitoring using lusu/cse721/papers/iSleep Unobtrusive Sleep... · PDF fileiSleep: Unobtrusive Sleep Quality Monitoring using Smartphones Tian Hao

Sep 07, 2018

ReportDownload

Documents

lynga

  • iSleep: Unobtrusive Sleep Quality Monitoring usingSmartphones

    Tian HaoMichigan State University

    Guoliang XingMichigan State University

    Gang ZhouCollege of William and Mary

    ABSTRACTThe quality of sleep is an important factor in maintain-ing a healthy life style. To date, technology has not en-abled personalized, in-place sleep quality monitoring andanalysis. Current sleep monitoring systems are often dif-ficult to use and hence limited to sleep clinics, or invasiveto users, e.g., requiring users to wear a device during sleep.This paper presents iSleep a practical system to moni-tor an individuals sleep quality using off-the-shelf smart-phone. iSleep uses the built-in microphone of the smart-phone to detect the events that are closely related to sleepquality, including body movement, couch and snore, and in-fers quantitative measures of sleep quality. iSleep adoptsa lightweight decision-tree-based algorithm to classify vari-ous events based on carefully selected acoustic features, andtracks the dynamic ambient noise characteristics to improvethe robustness of classification. We have evaluated iSleepbased on the experiment that involves 7 participants andtotal 51 nights of sleep, as well the data collected from realiSleep users. Our results show that iSleep achieves consis-tently above 90% accuracy for event classification in a va-riety of different settings. By providing a fine-grained sleepprofile that depicts details of sleep-related events, iSleep al-lows the user to track the sleep efficiency over time and relateirregular sleep patterns to possible causes.

    1. INTRODUCTIONSleep plays an important role in our overall health. Hav-

    ing insufficient amount of sleep can easily cause fatigue andlack of concentration during the day. Besides the amountof sleep, the quality of sleep is also an important factor inmaintaining a healthy life style. Clinical studies show thatsleep is related to many serious diseases including diabetes,obesity and depression [16] [27].

    This work is supported in part by the NSF undergrant CNS-0954039 (CAREER), CNS-1250180 and ECCS-0901437. This study is approved by the Institutional ReviewBoard (IRB) of Michigan State University.

    Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.SenSys13, November 1115, 2013, Rome, Italy.Copyright 2013 ACM 978-1-4503-1169-4 ...$15.00.

    To date, technology has not enabled personalized, in-placesleep quality monitoring and analysis. Polysomnography(PSG) is the primary clinical tool for sleep monitoring [13].It can provide a quantitative profiling of sleep to diagnosesleep disorders. However, due to the need of various sensors,PSG-based sleep quality measurement is usually limited toclinical settings. Actigraphy has been studied as an inex-pensive alternative to assess sleep and wakefulness based onbody movement [8]. Several portable sleep assessment prod-ucts are designed based on PSG or actigraphy technologies,including ZEO [7], Sleep Tracker [5] and fitbit [1]. However,they are invasive to users as they require a device to be wornby the user during sleep. A recent in-depth survey of 230participants suggested that, although most people are inter-ested in using technology to track their sleep quality, manyare resistant to the idea of having to wear a device duringsleep [14].

    This paper presents iSleep a practical system to mon-itor an individuals sleep quality using off-the-shelf smart-phone. iSleep is very easy to use and truly unobtrusive:the user just needs to start iSleep app and place the phonesomewhere close to the bed (e.g., on a night stand). iSleepuses the built-in microphone of the smartphone to detectthe events that are closely related to sleep quality, includingbody movement, couch and snore. Based on the detectedevents, iSleep infers quantitative measures of sleep qualitybased on actigraphy and Pittsburgh Sleep Quality Index (P-SQI) [12] which are two well-established scoring criteria insleep literature. We have released an initial version of iSleepon the Google Play Store [2]. Within 6 days, iSleep was in-stalled by more than 100 users from 9 countries on variousAndroid devices. By providing a detailed sleeping profile,iSleep enables the user to be aware of irregular sleep patternslike restlessness caused by extensive snoring which are oth-erwise hard to find. Moreover, as an unobtrusive, portable,in place monitoring tool, iSleep can track sleep quality quan-titatively over a long period of time, which helps healthcareprovider diagnose trends related to certain diseases.

    The design of iSleep faces several challenges such as highlydiverse acoustic profiles of sleep from person to person andin different environments. We carefully analyze the acousticdata collected from real sleep experiments and choose sever-al statistical acoustic features that can differentiate environ-ment noise and various sleep-related events. To improve therobustness of detection, iSleep tracks the ambient noise char-acteristics and updates the noise model adaptively. Finally,iSleep adopts a lightweight decision-tree-based algorithm toclassify various sleep-related events and derive quantitative

  • sleep quality measures. We have evaluated iSleep extensivelyin a long-term experiment that involves 7 participants andtotal 51 nights of sleep, as well as using the data collect-ed from the Android phones that downloaded and installediSleep from Google Play Store. Our results show that iSleepachieves consistently above 90% classification accuracy forvarious events, across different subjects and in a variety ofdifferent sleep environments.

    2. RELATED WORKAccording to AASM (American Academy of Sleep Medicine),

    the sleep stage scoring based on polysomnography (PSG)has long been considered as the gold standard of sleepstudy [20]. A polysomnogram typically requires the record-ing of multiple channels including electroencephalography(EEG), electromyography (EMG), electrocardiography (ECG)or heart rate, respiratory effort, air flow, oxygen saturationand etc. [13]. The result of PSG includes a collection ofindices such as sleep onset latency, total sleep time and etc,which are considered together to infer the sleep quality. Dueto the need of various sensors, PSG-based sleep quality mea-surement is usually limited to sleep clinics.

    Actigraphy has been studied as an inexpensive alternativeto assess human sleep and wakefulness [8] based on the sub-jects body movements overnight. The basic idea is that thestate of sleep and wake can be inferred from the amountof body movement during sleep [8]. Through processing thelogged acceleration data, epoch-by-epoch (usually 30 secondor 1 minute) sleep/wake predictions are calculated. Severalalgorithms [18] [29] [15] have been proposed to derive sleepquality from actigraphy. The average accuracy of predictingsleep/wake state is around 90% (reported 88% in [15] and94-96% in [31]).

    A widely used subjective sleep quality assessment methodis through PSQI (Pittsburgh Sleep Quality Index) [12], whichis a self-rated questionnaire to assess the sleep quality anddisturbance over a long-term interval. In PSQI, a set ofsleep measures are collected, including sleep latency, sleepduration, sleep disturbance and etc. PSQI has been shownuseful in numerous studies [9] [11] over a variety of popula-tions. However, the accuracy of PSQI is highly variable andis often impeded by the inaccuracy of subjects memory andperception.

    Several commercial personal sleep assessment productsare currently available. Watch PAT [6] detects respiratorydisturbances during sleep by monitoring peripheral arteri-al tone (PAT). The users are required to attach a probe totheir finger during sleep. ZEO [7] is a popular sleep monitor-ing product that infers sleep stages using three EEG sensorscontained in a head band worn by the user during sleep.Several actigraphy-based products such as Sleep Tracker [5]and fitbit [1] require the user to wear the device containingaccelerometer during sleep.

    Recently, several research efforts aimed at developing low-cost sleep assessment systems. In [28], a wearable neck-cuffsystem for real-time sleep monitoring is designed based onoximetry sensor, microphone and accelerometer. Instead ofdirectly measuring the sleep, SleepMiner [10] predicts thesleep quality based on the users daily context informationsuch as sound, light, postures, and positions. In [19], thebody position and movements during sleep are monitoredusing accelerometers attached to bed mattress. A densepressure sensitive bedsheet for sleep posture monitoring is

    proposed in [23]. However, these systems incur nontrivialmonetary costs of hardware or professional installation.

    Several Android and iOS Apps such as Sleep as Android[3] and Sleep Cycle [4] can measure sleep quality. All of themexclusively rely on the actigraphy-based methods that mon-itor body movements overnight using smartphones. Howev-er, sleep-related events such as cough and snore can not bereliably detected based on acceleration. For example, snoreis the sound caused by the vibration of respiratory struc-tures while sleeping due to obstructed air movement, andis not necessarily associated with body motion. Moreover,since the motion data is collected through the built-in ac-celerometer, the phone must be put on the bed, which notonly is inconsistent with the habit of most users, but alsomay obstruct the individuals body movement.

    iSleep leverages the existing body of work on acoustic sig-nal processing (e.g. SoundSense [24] and StressSense [25]).