Big data analysis of cyclic alternating pattern during sleep using deep learning

Abstract

Sleep scoring has been of great interest since the invention of the polysomnography method, which enabled the recording of physiological signals overnight. With the surge in wearable devices in recent years, the topic of what is high-quality sleep, how can it be determined and how can it be achieved attracted increasing interest. In the last two decades, cyclic alternating pattern (CAP) was introduced as a scoring alternative to traditional sleep staging. CAP is known as a synonym for sleep microstructure and describes sleep instability. Manual CAP scoring performed by sleep experts is a very exhausting and time-consuming task. Hence, an automatic method would facilitate the processing of sleep data and provide a valuable tool to enhance the understanding of the role of CAP. This thesis aims to expand the knowledge about CAP by developing a high-performance automated CAP scoring system that can reliably detect and classify CAP events in sleep recordings. The automated system is equipped with state-of-the-art signal processing methods and exploits the dynamic, temporal information in brain activity using deep learning. The automated scoring system is validated using large community-based cohort studies and comparing the output to verified values in the literature. Our findings present novel clinical results on the relationship between CAP and age, gender, subjective sleep quality, and sleep disorders demonstrating that automated CAP analysis of large population based studies can lead to new findings on CAP and its subcomponents. Next, we study the relationship between CAP and behavioural, cognitive, and quality-of-life measures and the effect of adenotonsillectomy on CAP in children with obstructive sleep apnoea as the link between CAP and cognitive functioning in children is largely unknown. Finally, we investigate cortical-cardiovascular interactions during CAP to gain novel insights into the causal relationships between cortical and cardiovascular activity that are underpinning the microstructure of sleep. In summary, the research outcomes in this thesis outline the importance of a fully automated end-to-end CAP scoring solution for future studies on sleep microstructure. Furthermore, we present novel critical information for a better understanding of CAP and obtain first evidence on physiological network dynamics between the central nervous system and the cardiovascular system during CAP.