Context-aware algorithms for sleep apnea monitoring and sensor acceptance using unobtrusive pressure sensors arrays

As older adults require more health care services, solutions are needed to shift the delivery of services from hospitals and nursing facilities to clients' homes. Many sleep disorders are serious enough to interfere with activities of daily living and have significant health and social consequences. Polysomnography, while invasive and expensive, provides a quantitative assessment of sleep quality by recording physiological variables with the use of multiple face and body electrodes.The goal of this thesis is to develop unobtrusive sleep monitoring technologies to facilitate non-invasive and home-based biomedical and preventive monitoring. This thesis presents multiple algorithms for use with unobtrusive pressure sensors in sleep monitoring. The algorithms for rollover detection, central apnea detection, sensor fusion, and delay profiling were validated with controlled experiments or with data acquired using polysomnography and unobtrusive sensors simultaneously. These algorithms incorporate context from patient-driven and data-driven sources to optimize parameters and screen patients. This thesis also presents a model of sensor acceptance as a tradeoff between privacy and autonomy.Algorithms were developed to address the importance of central apnea and the effect of positional sleeping on apnea severity. Using data collected from the West Ottawa Sleep Center, an algorithm classified central apnea events with a sensitivity >87%, specificity >99%, and a Cohen's kappa value of 0.875, which compares well with the current literature. The decision tree created using controlled experiment data identified rollovers with a sensitivity >81% and 100% specificity, which compliments existing static position detection schemes. A hypothesis was put forth that most older adults would likely not accept video surveillance in any situation, however frail older adults might be willing to "trade in their privacy" if it meant preventing nursing home placement. The hypothesis was accepted following a literature review of smart home monitoring sensor preferences.