Analyzing Fetal Heart Rate by Models Based on Hierarchical Dirichlet Process
June 22, 2018
Light Engineering room 250
Advisor: Prof. Petar M. Djuric
In this thesis, we propose to investigate the application of non-parametric Bayesian methods for processing and classification of fetal heart rate (FHR) recordings. More specifically, we propose models that are used to differentiate between FHR recordings that are from fetuses with or without adverse outcomes. In our work, we rely on models based on hierarchical Dirichlet processes (HDP) and the Chinese restaurant process with finite capacity (CRFC). We first show the theoretical formulation and simulation results of CRFC models. From the simulation, we observed that CRFC models were able to capture the changes of data evolving over time. For real FHR data, we employed several non-parametric Bayesian models for classification. Specifically, two generative models were inferred from FHR recordings, with one representing healthy and the other unhealthy fetuses. The models were then used to classify new recordings and provide the probability of the fetus being healthy or unhealthy. We compared the classification performance of the HDP-based models with that of support vector machines (SVMs) on real data by cross- validation method, and concluded that the HDP models achieved better performance. We also demonstrated the use of mixture models based on CRFC for dynamic classification of the performance of FHR recordings in a real-time setting.