Multi-Task Gaussian Process Regression for the Detection of Sleep Cycles in Premature Infants

• Published in: ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) pp. 1341 - 1345
• Main Authors: Brugge, Nele Sophie, Grasshoff, Jan, Weigenand, Arne, Rostalski, Philipp
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.05.2022
• Subjects: Annotations; Bayesian modeling; Correlation; Gaussian process; Gaussian processes; Medical services; Pediatrics; preterm infant; regression; Rhythm; sleep cycling; Temperature distribution
• ISSN: 2379-190X
• DOI: 10.1109/ICASSP43922.2022.9747086

Preterm birth is the leading cause of infant mortality. Consequently, preterm infants require special attention and medical care, with sleep being a central element for the development of cognitive functions. Studies on neonatal sleep suggest that the pattern of their sleep stages is determined by an endogenous ultradian rhythm, superimposed by other rhythms and external influences. In this article, we propose the use of multi-task Gaussian process regression as a flexible nonparametric approach to analyze this kind of sleep data while incorporating prior knowledge, such as of correlations between signals, signal periodicity, information from manual annotations and certain other signal properties. As a result of the regression of heart and respiratory rate data of preterm infants, ultradian rhythms with a period of 58±5min could be extracted. Together with other model parameters, knowledge about the characteristics of ultradian rhythms potentially provides insights into the maturational and health status of the preterm infants. These, in turn, could be used to optimize the care of critically ill patients.