Rationale: Sleep consolidation and rest/activity rhythm are disturbed in people with dementia, presumably due to higher prevalence of sleep disorders or degeneration in the related brain areas. Wrist actigraphy has been extensively used for extracting sleep measures due to its convenience in long continuous monitoring of activity in natural settings. Despite extensive applications, analytic algorithms used for extracting sleep measures from actigraphy are mostly validated on healthy young population that might be different in physiology and clinical conditions compared to older adults [9, 10], and are prone to errors due to individual differences, study assumptions and day-by-day changes. The central hypothesis is that artificial intelligence (AI) can offer more robust techniques than conventional algorithms to compensate for these subjective factors with less assumptions, and define more descriptive sleep measures with higher accuracy. Our previous study showed that deeper AI models could increase detection accuracy of sleep/wakefulness periods by >30% compared to conventional algorithms in older adults. Therefore, the overarching goal of this study is to develop AI-powered algorithms to estimate actigraphy-based sleep measures that predict the risk of cognitive decline.
In this study, we expand the outcomes through the following aims to:
Aim 1: Develop a package of AI-based algorithms to extract sleep measures, including: circadian rhythms, sleep fragmentation and subject-specific nap periods from wrist actigraphy;
Aim2: Develop predictive models using the extracted sleep measures and AI models to identify healthy sleeping schemes associated with lower risk of dementia.
These models will be developed.
Significance: This study will leverage the sleep analyses of large Canadian studies on aging by providing a reliable AI platform that can be used by other researchers in the field, and an insight about healthy sleeping schemes with reduced risk of dementia.
