Machine Learning to Classify Cardiotocography for Fetal Hypoxia Detection.

Area of Science:

• Obstetrics and Gynecology
• Biomedical Engineering
• Neonatal Medicine

Background:

• Fetal hypoxia during labor can lead to severe complications like stillbirth and cerebral palsy.
• Cardiotocography (CTG) is standard for monitoring fetal well-being but suffers from subjective interpretation, delaying interventions.
• Machine learning (ML) offers potential for automated CTG analysis, but requires clinically relevant benchmarks.

Purpose of the Study:

• To develop and evaluate ML models for classifying CTG signals using Apgar scores as a clinically relevant surrogate for neonatal recovery.
• To investigate the efficacy of ML in distinguishing between normal and potentially hypoxic fetal states based on CTG data.

Main Methods:

• Signal processing techniques were used to preprocess 552 raw CTG recordings.
• Validated features and NICE guideline-specific CTG characteristics were extracted.
• ML classifiers were employed using 22 features to analyze CTG data against Apgar scores.

Main Results:

• ML models demonstrated capability in classifying CTG associated with low Apgar scores.
• Performance for the rarest, lowest Apgar scores requires additional data for improvement.
• The study highlights the need for external datasets to validate model generalizability.

Conclusions:

• Apgar scores provide a clinically relevant benchmark for ML-based CTG classification.
• Automated CTG analysis using ML and Apgar scores shows promise for early fetal hypoxia detection.
• Further research with larger, diverse datasets is needed to refine and validate the model for widespread clinical application.