Sleep and Posture Determination from Raw Data Accelerometery

| Activinsights News

The GENEActiv and the Activinsights Band have capabilities beyond measures of pure physical activity. How do they classify between sitting and standing when there is no physical movement or how can they differentiate time spent in bed from time spent asleep?

Activinsights’ wearables contain a triaxial accelerometer that collects raw, high resolution movement data, and can also be used as a positional sensor to track and classify posture changes. When inactive and stationary, the primary signal from the accelerometer is from the Earth’s gravitational field and can be used to determine the orientation of the device and therefore, position. Any change in orientation results in a positional signal regardless of whether the wearable is in a state of low or dynamic acceleration. When standing, with the arm down against the body, there is a specific orientation output for each of the 3 axes of the accelerometer. Once the arm is rotated, for example, there are a series of small acceleration forces applied to the wearable, accompanied by a change of orientation. These changes of orientation can be mapped to provide a profile of an individual’s sedentary behaviour. We call this concept the Sedentary Sphere®, which is widely published. Using this method, researchers can describe both posture and low intensity behaviours (such as housework), which are often not fully recognised with more traditional methods of measurement, despite accounting for the vast proportion of most of a person’s day. The use of such techniques moves us a step closer to understanding and classifying a wider range of activities and measuring the impact they have on individual and population health.

These behaviours can be visualised on a sphere to display the pattern of wrist movement and positions associated with that behaviour. Arm elevation is plotted between the poles and wrist rotation around the equator. Often, there may be more than one sedentary posture associated with a specific sedentary behaviour. For example, when working at a desk, there will be clusters of postures associated with computer-based work, which are different and distinct to the clusters associated with time spent on the phone.


Seated office work

Watching television

The output from a night’s sleep can determine the number and frequency of postural changes and detect if an individual got out of bed during the night. Collecting sleep data in an individual’s own home environment results in a more realistic richer, true-to-life dataset than that collected in a laboratory setting. Whilst an accelerometer can detect and measure very fine resolution and low acceleration movements, the assessment of sleep is based on inference rather than direct measurement. For most participant populations however, the patterns of low movement are adequate to identify specific sleep events and the timings of the day-night cycle.

Many analysis methodologies can be applied to sleep data to help remove risk of misclassification. Commonly used are thresholding, change-point detection analysis, random forest classifiers and neural networks.

Employing objective measures of lifestyle removes the risk of self-report bias. Postural data analysis can provide a composite evaluation of an individual’s behaviours and inform lifestyle change or intervention. There is application in remote health monitoring to allow continuous recording of patient data. Coupling machine-learning with applying specific algorithms delivers a tool that can measure patient outcomes and increase understanding of health behaviours.

For further in-depth information on this topic, see references listed below.


Posture References:

Assessing sedentary behavior with the GENEActiv: introducing the sedentary sphere. Rowlands AV, Olds TS, Hillsdon M, Pulsford R, Hurst TL, Eston RG, Gomersall SR, Johnston K, Langford J. (2014) Med Sci Sports Exerc.  Jun;46(6):1235-47.

Fast and Robust Algorithm for Detecting Body Posture Using Wrist-Worn Accelerometers, Journal for the Measurement of Physical Behaviour. Straczkiewicz, M., Glynn, N. W., Zipunnikov, V., & Harezlak, J. (2020) Journals-HumanKinetics. 3(4), 285-293.

The validity of the GENEActiv wrist-worn accelerometer for measuring adult sedentary time in free living. Toby G. Pavey, Sjaan R. Gomersall, Bronwyn K. Clark, Wendy J. Brown. (2016) Science Direct. Volume 19, Issue 5,2016, Pages 395-399.

Wrist-Worn Accelerometer-Brand Independent Posture Classification. Rowlands AV, Yates T, Olds TS, Davies M, Khunti K, Edwardson CL. (2016) PubMed Abstract. Apr;48(4):748-54.

Accuracy of Posture Allocation Algorithms for Thigh- and Waist-Worn Accelerometers. Edwardson CL, Rowlands AV, Bunnewell S, Sanders J, Esliger DW, Gorely T, O’Connell S, Davies MJ, Khunti K, Yates T. (2016) PubMed Abstract. Jun;48(6):1085-90.

Validating the Sedentary Sphere method in children: Does wrist or accelerometer brand matter? Liezel Hurter, Alex V. Rowlands, Stuart J. Fairclough, Karl C. Gibbon, Zoe R. Knowles, Lorna A. Porcellato, Anna M. Cooper-Ryan & Lynne M. Boddy. (2019) Taylor & Francis Online Abstract. 37:16, 1910 1918.

Sleep References:

Sleep Estimates Using Microelectromechanical Systems (MEMS). Bart H. W. te Lindert, MSc, Eus J. W. Van Someren, PhD. (2013) Academic Oup. Volume 36, Issue 5, 1 May 2013, Pages 781–789.

A Novel, Open Access Method to Assess Sleep Duration Using a Wrist-Worn Accelerometer. Van Hees VT, Sabia S, Anderson KN, Denton SJ, Oliver J, Catt M, et al. (2015) Journals Plos. PLoS ONE 10(11): e0142533.

Estimating sleep parameters using an accelerometer without sleep diary. Van Hees, V.T., Sabia, S., Jones, S.E. et al. (2018) Nature. Sci Rep 8, 12975

“Analyzing sensor based human activity data using time series segmentation to determine sleep duration”. Lad, Yogesh Deepak. (2018) Core AC. Masters Theses. 7802.

Objective Characterization of Activity, Sleep, and Circadian Rhythm Patterns Using a Wrist-Worn Actigraphy Sensor: Insights Into Posttraumatic Stress Disorder. Tsanas A, Woodward E, Ehlers A. (2020) Ncbi. 8(4):e14306.

Equivalency of Sleep Estimates: Comparison of Three Research-Grade Accelerometers, Journal for the Measurement of Physical Behaviour. Plekhanova, T., Rowlands, A. V., Yates, T., Hall, A., Brady, E. M., Davies, M., Khunti, K., & Edwardson, C. L. (2020) Journals Human-Kinetics. 3(4), 294-303.

Objective Sleep Duration in Older Adults: Results from The Irish Longitudinal Study on Ageing. Siobhán Scarlett, MSc, Hugh Nolan, PhD, Rose Anne Kenny, MD, and Matthew DL O’Connell, PhD. (2020) AGS Journals.

Device agnostic sleep-wake segment classification from wrist-worn accelerometry. Luis R. Peraza, Richard Joules, Yves Dauvilliers, Robin Wolz. (2020) Ixico.