Automated assessment of the first trimester placental volume and uterine vasculature using three-dimensional power Doppler ultrasound to predict adverse perinatal outcomes

Thesis

Placental insufficiency syndromes, including fetal growth restriction (FGR) and pre-eclampsia (PE), are leading causes of perinatal and maternal morbidity and mortality. Early identification of at-risk pregnancies remains challenging due to limitations in current screening and the complexity of placental biology. This thesis evaluates three-dimensional power Doppler ultrasound, processed using machine learning, for quantifying first trimester placental volume and uterine vasculature with the goal of improving prediction of adverse perinatal outcomes.

Data from the FirstPLUS study were used as a reference population and for model development. The OxPLUS cohort was prospectively established to develop and externally validate novel placental imaging biomarkers and predictive models for placental insufficiency. The OxNNet deep learning framework enabled rapid, reproducible segmentation of the placenta and automated calculation of single vessel fractional moving blood volume (svFMBV).

Prescriptive centile charts for placental volume and svFMBV were constructed, and derived z-scores were significantly associated with FGR and PE. Multivariable models incorporating OxNNet-derived metrics, maternal history, and established biomarkers demonstrated good discrimination for both conditions, which was attenuated but preserved when the models were assessed in the independent OxPLUS cohort. The centile charts themselves have not yet been assessed in an independent population.

This thesis also introduces new quantitative markers of placental and vascular morphology, several of which showed significant associations with adverse outcomes. The feasibility of training sonographers in advanced placental ultrasound was demonstrated, alongside the implementation of structured quality assurance tools, resulting in good reproducibility for most OxNNet-derived metrics.

These findings establish proof of concept for automated placental assessment in early pregnancy and provide a foundation for further research into placental insufficiency. This work advances the quantitative assessment of placental function, but further validation in independent populations, together with assessment of calibration and of clinical impact, would be required before it could inform obstetric practice.

Authors: Mathewlynn, S

• DOI: 10.5287/ora-2rv47d4jk
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: pre-eclampsia; placental vascularity; fetal growth restriction; prenatal screening; deep learning
• Subjects: Prenatal diagnosis; Placenta; Doppler ultrasonography; Ultrasonics in obstetrics; Obstetrics; Machine learning