Assessment of features for automatic CTG analysis based on expert annotation

Area of Science:

• Obstetrics and Gynecology
• Biomedical Engineering
• Fetal Medicine

Background:

• Cardiotocography (CTG) has monitored fetal heart rate (FHR) and uterine contractions (TOCO) since the 1960s for fetal hypoxia detection.
• Current FHR evaluation relies on macroscopic features, largely ignoring advances in heart rate variability (HRV) research.

Purpose of the Study:

• To investigate the statistical significance of various FHR features for classifying FHR into three FIGO classes.
• To identify the most effective and uncorrelated FHR features for clinical application.

Main Methods:

• Utilized a large dataset of 552 CTG records with expert annotations.
• Assessed a comprehensive set of FHR features, including FIGO, HRV, nonlinear, wavelet, and time/frequency domain features.
• Employed meta-analysis of three ranking methods to determine feature importance.

Main Results:

• Identified key features with statistical significance for FHR classification.
• Number of accelerations/decelerations, interval index, Lempel-Ziv complexity, and Higuchi's fractal dimension ranked among the top five features.
• Established a ranked list of uncorrelated, important FHR features.

Conclusions:

• Several FHR features demonstrate significant potential for improving fetal hypoxia detection.
• The study provides a robust feature selection for enhanced FHR analysis in clinical practice.
• Highlights the value of incorporating advanced HRV and complexity measures into routine CTG analysis.