Cardiotocography analysis by empirical dynamic modeling and Gaussian processes

Area of Science:

• Medical Informatics
• Dynamical Systems Theory
• Machine Learning

Background:

• Cardiotocography (CTG) is standard for electronic fetal monitoring, but interpretation has high variability.
• Existing machine learning for CTG analysis often overlooks uterine activity (UA) signals.
• Obstetricians evaluate both fetal heart rate (FHR) and UA for fetal well-being assessment.

Purpose of the Study:

• To model intrapartum CTG recordings using empirical dynamic modeling with Gaussian processes.
• To investigate the causal relationship between FHR and UA signals.
• To leverage these models for enhanced computerized analysis of CTG data.

Main Methods:

• Utilized Gaussian processes, a Bayesian nonparametric approach, for function estimation.
• Modeled CTG recordings from a dynamical system perspective.
• Applied empirical dynamic modeling to time series data of FHR and UA.

Main Results:

• Gaussian processes enabled simultaneous estimation of attractor manifold dimensionality and reconstruction.
• Demonstrated a causal relationship between FHR and UA signals in real CTG recordings.
• Developed models capable of estimating missing FHR samples and recovering burst errors.

Conclusions:

• The causal relationship between FHR and UA is significant for fetal well-being assessment.
• Gaussian process modeling offers a robust framework for analyzing complex CTG data.
• This approach has potential applications in improving the accuracy and reliability of electronic fetal monitoring.