EXTRACTING INTERPRETABLE FEATURES FOR FETAL HEART RATE RECORDINGS WITH GAUSSIAN PROCESSES.

Area of Science:

• Obstetrics and Gynecology
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Cardiotocography (CTG) monitors fetal heart rate (FHR) and uterine activity (UA) during labor.
• Visual inspection of CTG by obstetricians suffers from high inter- and intra-variability.
• Current computerized FHR analysis uses interpretable features with low correlation to umbilical cord blood pH or less interpretable machine learning features.

Purpose of the Study:

• To explore the potential of Gaussian process (GP) hyperparameters as interpretable features for FHR analysis.
• To improve the correlation between FHR features and umbilical cord blood pH, the gold standard for fetal well-being assessment.
• To develop FHR analysis methods that balance interpretability with predictive accuracy.

Main Methods:

• Utilized Gaussian processes (GPs) for the analysis of FHR signals.
• Investigated the use of GP hyperparameters as novel, interpretable features.
• Correlated the derived GP features with umbilical cord blood pH values.
• Assessed the correlation of GP features with existing popular FHR features.

Main Results:

• Identified specific GP features that demonstrate a high correlation with umbilical cord blood pH.
• Found that these novel GP features are not highly correlated with commonly used FHR features.
• Demonstrated the potential of GP hyperparameters to serve as interpretable and clinically relevant indicators of fetal well-being.

Conclusions:

• Gaussian process hyperparameters can serve as effective and interpretable features for FHR analysis.
• This approach offers a promising alternative to existing methods, enhancing the assessment of fetal well-being during labor.
• Further research into GP-based features could lead to more objective and reliable CTG interpretation.