Classification of intrauterine growth restriction at 34-38 weeks gestation with machine learning models

Area of Science:

• Perinatal Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Intrauterine growth restriction (IUGR) is a significant risk factor for stillbirth.
• Early identification of IUGR is crucial for timely intervention and improved perinatal outcomes.
• Predictive modeling offers a novel approach to identify fetuses at risk for IUGR.

Purpose of the Study:

• To develop and validate a machine learning model for predicting IUGR.
• To identify fetuses with estimated fetal weight (EFW) below the 10th percentile.
• To utilize data from early gestation (20-24 weeks) for prediction at a later stage (34-38 weeks).

Main Methods:

• Prospective data collection from the Safe Passage Study (SPS) over 7.5 years.
• Inclusion of maternal/fetal ECG and ultrasound biometry/Doppler data.
• Application of supervised learning techniques including Stochastic Gradient Descent, k-NN, Logistic Regression, and Random Forest.

Main Results:

• The developed model achieved high performance across metrics (93% accuracy, precision, recall, F1-score).
• The Umbilical Artery Pulsatility Index was identified as the strongest predictor of IUGR.
• ROC analysis demonstrated a strong True Positive rate with an AUC of 0.771.

Conclusions:

• The machine learning model demonstrates robust and flexible predictive capabilities for IUGR.
• Pre-processed features, including fetal heart rate variability, enhanced model accuracy.
• The model provides a foundation for predicting other birth-related anomalies and developing more complex predictive tools.