Prediction of Intrauterine Growth Restriction and Preeclampsia Using Machine Learning-Based Algorithms: A Prospective...

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Computational Biology

Background:

• Screening for intrauterine growth restriction (IUGR) and preeclampsia (PE) presents a significant challenge for prenatal care providers.
• Accurate early detection of PE and IUGR is crucial for improving maternal and fetal outcomes.
• Existing screening methods often have limitations in predicting these complex pregnancy disorders.

Purpose of the Study:

• To assess and compare the predictive accuracy of four machine learning algorithms for PE and IUGR.
• To evaluate the performance of Decision Tree (DT), Naïve Bayes (NB), Support Vector Machine (SVM), and Random Forest (RF) models.
• To investigate the prediction of PE, IUGR, and their combined occurrence using clinical and paraclinical data.

Main Methods:

• An observational prospective study involving 210 singleton pregnancies.
• First-trimester screening data, including clinical and paraclinical information, were utilized.
• Four machine learning algorithms (DT, NB, SVM, RF) were employed to predict PE and IUGR.

Main Results:

• The Random Forest (RF) algorithm achieved the highest accuracy: 96.3% for PE, 95.9% for IUGR, and 95.1% for their association.
• RF also demonstrated high accuracy for early-onset IUGR (96.2%) and late-onset IUGR (95.2%).
• SVM and NB showed comparable IUGR prediction (95.3%), while SVM slightly outperformed NB in PE prediction (95.8% vs. 94.7%).

Conclusions:

• Machine learning algorithms, particularly Random Forest, show significant potential for improving the early detection of PE and IUGR.
• Integration of these algorithms into first-trimester screening could enhance detection rates for these critical obstetric conditions.
• Further validation in larger, diverse cohorts is recommended to confirm these findings and facilitate clinical implementation.