Graduate Computer and Informa^on Sciences PhD Abstract ID: 492 ...
Graduate Computer and Informa0on Sciences PhD Abstract ID: 492
A System For Gathering Data on Sleep Behavior and Context in the Home SeBng Aida Ehyaei¹, Stephen In0lle²
¹Department of Electrical and Computer Engineering, Northeastern University ²College of Computer and Informa0on Science and Bouvé College of Health Sciences, Northeastern University A collabora0on with Case Western Reserve University Problem We spend almost one-‐third of our lives sleeping. Quality and quan0ty of sleep affect health and how people feel and behave during the day, but rela0vely liNle is known about how home behavior impacts sleep quality. Therefore, beNer tools are needed to study sleep behaviors unobtrusively, outside of the laboratory. Background • Polysomnography, the gold standard laboratory sleep monitoring method is expensive and burdensome for the person wearing the system; typically the electrodes required must be aNached by an expert. • Ac6graphy, where accelerometers measure a person’s wrist movement, is an inexpensive method of measuring sleep but does not provide informa0on about a person’s sleep environment that may impact sleep quality, such as noise. Approach We developed a system for sleep monitoring in home seRngs using context-‐sensi0ve self-‐report. The system includes a mo0on sensor, an applica0on on an Android based smartphone, and a remote data collec0on and visualiza0on system. Ac0graphy and audio amplitude analysis is used to trigger self-‐report and gather data on not only sleep quality, but also sleep environment and possible sleep disruptors. Data Gathering • Mo6on data are gathered using a wrist worn accelerometer sensor, Wocket, and sent wirelessly, in real-‐0me, to the assigned smart phone, leT on the bedside table. • Ambient noise levels are also monitored using the phone’s microphone.
Phone Applica6on • Real-‐6me processing of incoming mo0on and audio data, inferring sleep/wake states and also marking poten0ally disrup0ve audio events throughout the night. • Execu0ng a context-‐aware and 0me-‐dependent adap6ve survey (Fig. 2) with tailored ques0ons about the events of the night to elicit more precise answers from the user about poten0al sleep disturbances. • Sending data to a remote study data monitoring and visualiza0on tool. Online data visualizer The mobile system simplifies research study administra0on by sending data to a research webserver regularly for incremental data cleaning and compliance checking. (Fig. 3)
Fig. 3: Server-‐side soTware allows the research team to monitor the deployed systems hourly, iden0fying problems quickly and inspec0ng the mo0on and audio data for each par0cipant.
Sleep-‐wake scoring algorithms To validate sleep detec0on algorithm from Wocket data we compared the output from our system to sleep states determined by polysomnography on 10 par0cipants. We implemented three algorithms used for detec0ng sleep vs. wake states from accelera0on data in the sleep literature. Table1: The accuracy of sleep/wake detec0on algorithms for Wocket mo0on data using three different algorithms on ten people for one day of data. Sensi6vity Specificity
Accuracy
(sen + spe)
0.71
Sadeh Algorithm [1] 0.87 0.83
1.58
0.82
Kripke Algorithm [2] 0.82 0.82
1.64
0.76
Sazonov Algorithm [3] 0.85 0.82
1.60
Conclusion Using mo0on sensors and smart phones allows the measurement of sleep in home seRngs for several nights. The system is currently being pilot tested in the homes of children in Cleveland, OH and will be used to assess sleep disturbances and peer and family effects on urban African-‐American children’s sleep. Acknowledgements This is joint work with James Spilsbury at Case Western Reserve University and funded by NIH/NIMHD grant #R21 MD007632. Funding for some of the tools that were used or extended for this work was made possible by NIH/NHLBI grant #U01HL091737. Fig. 2: Survey ques0ons are prompted on the phone before study par0cipants go to bed and aTer the system detects they have woken up; this context-‐ sensi0ve self report is and used to gather informa0on on events in the home that may impact sleep onset and quality.
References Fig. 1: A Wocket, the custom wearable sensors used in this work; and the components in a Wocket “kit” given to study par0cipants include 2 Wockets, a charger, and a comfortable band.
[1]Sadeh, A. et al. 1995. Ac0vity-‐based assessment of sleep-‐wake paNerns during the 1st year of life. Infant Behavior and Development. 18, 3 (1995), 329–337. [2]Kripke, D.F. et al. 2010. Wrist ac0graphic scoring for sleep laboratory pa0ents: algorithm development: Wrist ac0graphic algorithm development. Journal of Sleep Research. 19, 4 (Dec. 2010), 612–619. [3]Sazonov, E. et al. 2004. Ac0vity-‐based sleep–wake iden0fica0on in infants. Physiological Measurement. 25, 5 (Oct. 2004), 1291–1304.
A System For Gathering Data on Sleep Behavior and Context in the Home SeBng Aida Ehyaei¹, Stephen In0lle²
¹Department of Electrical and Computer Engineering, Northeastern University ²College of Computer and Informa0on Science and Bouvé College of Health Sciences, Northeastern University A collabora0on with Case Western Reserve University Problem We spend almost one-‐third of our lives sleeping. Quality and quan0ty of sleep affect health and how people feel and behave during the day, but rela0vely liNle is known about how home behavior impacts sleep quality. Therefore, beNer tools are needed to study sleep behaviors unobtrusively, outside of the laboratory. Background • Polysomnography, the gold standard laboratory sleep monitoring method is expensive and burdensome for the person wearing the system; typically the electrodes required must be aNached by an expert. • Ac6graphy, where accelerometers measure a person’s wrist movement, is an inexpensive method of measuring sleep but does not provide informa0on about a person’s sleep environment that may impact sleep quality, such as noise. Approach We developed a system for sleep monitoring in home seRngs using context-‐sensi0ve self-‐report. The system includes a mo0on sensor, an applica0on on an Android based smartphone, and a remote data collec0on and visualiza0on system. Ac0graphy and audio amplitude analysis is used to trigger self-‐report and gather data on not only sleep quality, but also sleep environment and possible sleep disruptors. Data Gathering • Mo6on data are gathered using a wrist worn accelerometer sensor, Wocket, and sent wirelessly, in real-‐0me, to the assigned smart phone, leT on the bedside table. • Ambient noise levels are also monitored using the phone’s microphone.
Phone Applica6on • Real-‐6me processing of incoming mo0on and audio data, inferring sleep/wake states and also marking poten0ally disrup0ve audio events throughout the night. • Execu0ng a context-‐aware and 0me-‐dependent adap6ve survey (Fig. 2) with tailored ques0ons about the events of the night to elicit more precise answers from the user about poten0al sleep disturbances. • Sending data to a remote study data monitoring and visualiza0on tool. Online data visualizer The mobile system simplifies research study administra0on by sending data to a research webserver regularly for incremental data cleaning and compliance checking. (Fig. 3)
Fig. 3: Server-‐side soTware allows the research team to monitor the deployed systems hourly, iden0fying problems quickly and inspec0ng the mo0on and audio data for each par0cipant.
Sleep-‐wake scoring algorithms To validate sleep detec0on algorithm from Wocket data we compared the output from our system to sleep states determined by polysomnography on 10 par0cipants. We implemented three algorithms used for detec0ng sleep vs. wake states from accelera0on data in the sleep literature. Table1: The accuracy of sleep/wake detec0on algorithms for Wocket mo0on data using three different algorithms on ten people for one day of data. Sensi6vity Specificity
Accuracy
(sen + spe)
0.71
Sadeh Algorithm [1] 0.87 0.83
1.58
0.82
Kripke Algorithm [2] 0.82 0.82
1.64
0.76
Sazonov Algorithm [3] 0.85 0.82
1.60
Conclusion Using mo0on sensors and smart phones allows the measurement of sleep in home seRngs for several nights. The system is currently being pilot tested in the homes of children in Cleveland, OH and will be used to assess sleep disturbances and peer and family effects on urban African-‐American children’s sleep. Acknowledgements This is joint work with James Spilsbury at Case Western Reserve University and funded by NIH/NIMHD grant #R21 MD007632. Funding for some of the tools that were used or extended for this work was made possible by NIH/NHLBI grant #U01HL091737. Fig. 2: Survey ques0ons are prompted on the phone before study par0cipants go to bed and aTer the system detects they have woken up; this context-‐ sensi0ve self report is and used to gather informa0on on events in the home that may impact sleep onset and quality.
References Fig. 1: A Wocket, the custom wearable sensors used in this work; and the components in a Wocket “kit” given to study par0cipants include 2 Wockets, a charger, and a comfortable band.
[1]Sadeh, A. et al. 1995. Ac0vity-‐based assessment of sleep-‐wake paNerns during the 1st year of life. Infant Behavior and Development. 18, 3 (1995), 329–337. [2]Kripke, D.F. et al. 2010. Wrist ac0graphic scoring for sleep laboratory pa0ents: algorithm development: Wrist ac0graphic algorithm development. Journal of Sleep Research. 19, 4 (Dec. 2010), 612–619. [3]Sazonov, E. et al. 2004. Ac0vity-‐based sleep–wake iden0fica0on in infants. Physiological Measurement. 25, 5 (Oct. 2004), 1291–1304.
