Abstract
BACKGROUND: Volumetric assessment of the fetus, placenta and amniotic fluid is clinically valuable, but MRI volumetry is rarely performed in clinical practice because of the required labour-intensive manual segmentation of motion-corrupted 2-dimensional (2-D) stacks. Existing deep-learning approaches typically segment single structures in 2-D motion-corrupted stacks, are, however limited in accuracy by slice misalignment. No current method provides a reliable automated solution for whole-uterus volumetry in 3-D reconstructed MRI. Furthermore, normative ranges for computation of centiles are currently missing. OBJECTIVE: To develop an automated pipeline for whole-uterus volumetry in 3-D T2-weighted fetal MRI and to generate normative growth models for fetal, placental and amniotic fluid volumes in healthy pregnancies with confirmed delivery at term. MATERIALS AND METHODS: Deformable slice-to-volume 3-D reconstruction was applied to motion-corrupted T2-weighted (T2W) stacks from 0.55T-3T MRI, and a 3-D UNet was trained to segment fetus, placenta and amniotic fluid on the resulting reconstructed 3-D images. A reporting tool generates centiles, z-scores and structured HTML outputs. Automated segmentation was performed in 357 healthy control datasets from 16-41 weeks gestational age (GA) range with confirmed delivery at term. After visual checks of segmeted labels and minor refinements, GA-based quadratic normative volumetry models were derived and correlations with maternal and fetal characteristics assessed. The utility of the pipeline for clinical research was further evaluated using 95 longitudinal scans from 42 fetuses and 86 preterm (≤ 32 weeks at delivery) pregnancies. RESULTS: Automated segmentation produced accurate 3-D labels, with only small local corrections (< 1% volume difference) required in the control cohort(< 25% of the datasets). Fetal and placental volumes increased across gestation, while amniotic fluid volume peaked mid-pregnancy and declined toward term. Volumes and centiles correlated with maternal size and birth weight. Longitudinal scans showed individual fetal and placental trajectories closely following the normative curves, with greater variability in amniotic fluid. Preterm pregnancies showed significantly lower fetal, placental and amniotic fluid volumes and centiles than the controls with confirmed delivery at term. CONCLUSION: This study introduces an automated whole-uterus volumetry pipeline and corresponding normative 3-D MRI growth models. The method provides robust, standardised volumetric assessment of fetal, placental and amniotic fluid development and offers a practical tool for evaluating growth patterns in both normal and high-risk pregnancies.