A Survey of Computerized Fetal Heart Rate Monitoring and Interpretation Techniques

Fetal heart rate (FHR) reveals crucial information about fetal conditions and FHR monitoring is a standard clinical procedure widely used during pregnancy and labor. FHR tracing can be interpreted as reassuring, non-reassuring, and ominous patterns. Accurate interpretation of fetal state from FHR tracing can prompt necessary interventions before irreversible conditions occur and reduce the number of unnecessary cesarean sections. Although FHR tracing can be obtained automatically using electrocardiograms and cardiotocography, FHR monitoring and interpretations are conducted manually by nurses or obstetricians, resulting in substantial inter-observer and intra-observer disparities. There have been great needs to automate FHR monitoring and interpretation in order to produce accurate and consistent clinical judgements and practices. A number of methods have been developed to address various issues in computerized automatic FHR monitoring and interpretation, including FHR signal modeling and representation, feature extraction, and pattern classification. FHR signals can be considered as stochastic processes driven by the cardiovascular control system and both linear and nonlinear methods have been proposed to model FHR signals. A wide range of parameters, such as morphological features, temporal domain parameters, frequency domain parameters, and nonlinear parameters, have been defined to represent different FHR features. Finally, various classification methods have been utilized to categorize FHR signals into reassuring, non-reassuring, and ominous patterns. This paper provides a comprehensive review of various approaches in FHR signal modeling, feature extraction, and pattern classification and discusses several methods that are under investigation and can possibly improve FHR modeling and classification. Index Terms — Fetal heart rate, modeling, feature extraction, spectrum analysis, pattern classification.