Fetal Heart Rate Analysis with Gaussian Processes

In this chapter, we present Gaussian process-based machine learning solutions to important tasks in computerized fetal heart rate analysis. Gaussian processes provide a powerful Bayesian machinery for learning functions or mappings and have inherent connections with many popular machine learning methods such as support vector machines and neural network models. More importantly, thanks to their Bayesian nature, Gaussian process-based models allow uncertainty quantification and data-efficient learning. This makes them suitable for solving learning tasks in perinatal medicine. The chapter is organized as follows. In Sect. 1, we provide a gentle introduction to Gaussian processes. In the next section, we introduce deep Gaussian processes. We describe an open access intrapartum CTG database that was used throughout all our experiments in Sect. 3. In Sect. 4, we present a Gaussian process-based approach to estimate missing samples of fetal heart rate recordings. In Sects. 5 and 6, we present a deep Gaussian process-based approach to classification and clustering of FHR recordings, respectively.