ANALYSIS OF FETAL HEART RATE SERIES BY NONPARAMETRIC HIDDEN MARKOV MODELS

Fetal heart rate (FHR) signals are routinely monitored to help obstetricians assess fetal status. In addition to guidelines for visual inspections, much research has been focused on computerized analysis of FHR tracings. In this paper, we propose to process FHR series by hidden Markov models (HMMs) and associate the hidden states with patterns of the tracings. Furthermore, we employ a nonparametric Bayesian approach, which does not define the number of hidden states beforehand, but instead uses data to determine the most appropriate number of states. We propose to use a nonparametric HMM, known as sticky hierarchical Dirichlet process-hidden Markov model (HDP-HMM) to resolve problems that arise due to redundant states and rapid switching rate of basic non parametric models. We use the HDP-HMMs to classify FHR signals into two groups and compare the results with those of support vector machines (SVMs). The classification performance showed that the HMM-based method achieved better accuracy.