Predicting gestational age at birth in the context of preterm birth from multi-modal fetal MRI

Preterm birth is a leading cause of neonatal mortality and long-term morbidity, yet its multifactorial nature makes accurate prediction difficult. Current clinical practice relies on limited indicators like cervical length and history of prior preterm birth. This study investigates whether a machine learning pipeline, integrating multi-modal fetal MRI (anatomical and functional T2* relaxometry) with clinical and ultrasound data, can improve the prediction of gestational age (GA) at birth.

The researchers developed a stacking-based machine learning pipeline to handle the complexities of the data, including missing values and feature imbalance. The pipeline incorporates data imputation, feature selection, and an ensemble of regression models (Random Forests, Support Vector Regression, and XGBoost) to predict GA at birth. The model was trained and evaluated using stratified 10-fold cross-validation on a cohort of 426 pregnancies (333 term, 93 preterm).

The pipeline achieved a mean absolute error of 2.74 weeks in predicting the exact gestational age at birth. When categorized into term and preterm outcomes, the model reached an accuracy of 0.77, with a sensitivity of 0.59 and a specificity of 0.82. The study identified that the most predictive features for the model were cervical length and statistics derived from placental T2* values, highlighting the utility of combining morphological and functional imaging markers.

This work serves as a proof of concept for using multi-modal fetal MRI to move beyond simple binary classification of preterm birth risk. By providing a continuous prediction of gestational age at birth, this approach could eventually offer more personalized clinical insights, potentially allowing for better risk stratification and management of pregnancies. The authors emphasize that this is a first step, with future work aimed at expanding the cohort size to refine predictions within the preterm population.