Fetal heart rate classification using generative models.

Area of Science:

• Medical signal processing
• Computational biology
• Bayesian inference

Background:

• Accurate classification of fetal heart rate (FHR) signals is crucial for clinical decision-making during labor.
• Current methods often rely on scalar features, potentially losing information from local FHR patterns.
• Existing classification systems include expert systems and support vector machines (SVMs) using heart rate variability (HRV) features.

Purpose of the Study:

• To develop and evaluate novel methods for classifying fetal heart rate (FHR) signals using generative models (GMs) and Bayesian theory.
• To enable the explicit use of feature sequences derived from local FHR evolution patterns.
• To compare the performance of the proposed methods against established classification techniques.

Main Methods:

• Development of classification methods based on generative models (GMs) and Bayesian inference.
• Utilizing feature sequences capturing local patterns of FHR evolution, rather than scalar summaries.
• Comparison with a deterministic expert system and a support vector machine (SVM) approach employing system-identification and HRV features.

Main Results:

• The proposed GM and Bayesian methods demonstrated performance comparable to or exceeding that of the expert system and SVM.
• The novel methods effectively utilized feature sequences for improved FHR signal classification.
• Consistent performance across 83 retrospectively collected FHR records validated the approach.

Conclusions:

• Generative models and the Bayesian paradigm offer significant improvements for automatic fetal heart rate (FHR) classification.
• The developed methods provide a more robust approach for clinical implementation of FHR analysis.
• Explicitly modeling feature sequences enhances the accuracy of FHR signal interpretation.