Comparative study of neonatal brain injury fetuses using machine learning methods for perinatal data

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Cardiotocography (CTG) is crucial for monitoring fetal well-being during pregnancy, assessing fetal heart rate and uterine contractions.
• Perinatal asphyxia, a major cause of neonatal hypoxic-ischemic encephalopathy, contributes significantly to neonatal death and disability.
• Effective prenatal monitoring is vital to reduce perinatal morbidity and mortality and prevent severe neurological damage.

Purpose of the Study:

• To evaluate the classification performance of various machine learning (ML) methods using CTG data.
• To assess the potential of ML models to assist clinicians in diagnosing fetal health status.
• To identify the most effective ML algorithm for analyzing CTG signals.

Main Methods:

• Utilized a dataset of 2126 CTG samples from the public UCI database.
• Applied and evaluated several machine learning algorithms including XGBoost, Random Tree, LightGBM, Decision Tree, and KNN.
• Assessed classification performance using metrics such as accuracy, precision, recall, and F1 score.

Main Results:

• All evaluated ML models achieved over 80% accuracy in classifying CTG data.
• XGBoost demonstrated the highest accuracy (91%) and superior performance across precision (90%), recall (93%), and F1 score (90%).
• Other models like Random Tree and LightGBM also showed strong performance, with accuracies of 90%.

Conclusions:

• Machine learning, particularly XGBoost, offers a feasible and effective approach for analyzing CTG data to assess fetal health.
• The high performance of ML models provides an objective tool to support clinical judgment and improve diagnostic accuracy.
• Implementing ML in CTG analysis can aid in early detection of fetal distress, potentially reducing adverse perinatal outcomes.