Enhanced multi-source data analysis for personalized sleep-wake pattern recognition and sleep parameter extraction

Abstract

Sleep behavior is traditionally monitored with polysomnography, and sleep stage patterns are a key marker for sleep quality used to detect anomalies and diagnose diseases. With the growing demand for personalized healthcare and the prevalence of the Internet of Things, there is a trend to use everyday technologies for sleep behavior analysis at home, having the potential to eliminate expensive in-hospital monitoring. In this paper, we conceived a multi-source data mining approach to personalized sleep-wake pattern recognition which uses physiological data and personal information to facilitate fine-grained detection. Physiological data includes actigraphy and heart rate variability and personal data makes use of gender, health status and race information which are known influence factors. Moreover, we developed a personalized sleep parameter extraction technique fused with the sleep-wake approach, achieving personalized instead of static thresholds for decision-making. Results show that the proposed approach improves the accuracy of sleep and wake stage recognition, therefore, offers a new solution for personalized sleep-based health monitoring.

Personalization of Automatic Sleep Scoring: How Best to Adapt Models to Personal Domains in Wearable EEG

“…This finding is reflected in automatic sleep scoring models: deep learning models have increased performance on individual subjects when the personal contribution to the domain is accounted for [9], [10]. In some approaches, this personal contribution is accounted for by extending the input space with metadata features [11]- [13]. In this study, we consider the personal contribution intrinsic to the EEG signal, and we investigate to which extent personalization, as adapting a sleep scoring model (using only EEG data) can increase sleep scoring performance on a single subject.…”