The Exploratory Role of AI-Assisted Modelling in the Assessment of Labour Progress: The Value of Ultrasound Parameters Compared to Clinical Evaluation

Abstract

This study analysed the potential of intrapartum ultrasound, enhanced by artificial intelligence (AI)-based interpretation, to optimise labour progress evaluation compared with standard clinical examination.
Methods: Two prospectively collected databases from Cuza Voda Clinical Hospital of Obstetrics and Gynecology were examined: one in the birth room (Eco in Birth Room, 2023-2024) and one during labour (Eco during Labour, 2021-2022). Ultrasound characteristics, namely the Head–Perineum Distance (HPD, cm), the Angle of Progression (AoP, °), the Head–Symphysis Distance (HSD, cm), and foetal station (clinically assessed) were retrieved and compared with clinical outcomes, particularly the mode of delivery (vaginal or caesarean). Data were analysed using R 4.4.2 using the packages tidyverse, readxl, janitor, and stats, logistic regression models were fitted with stats::glm(). No ultrasound images or comprehensive imaging datasets were utilized, and no segmentation algorithms were employed, as the model depended solely on manually acquired quantitative measurements. Predictive performance was measured using simple classification accuracy, which is based on the fraction of properly predicted delivery modes in the same dataset (no cross-validation). This accuracy metric is exploratory and meant to demonstrate the potential utility of AI-assisted quantitative modeling.The selection of logistic regression ensured enhanced transparency and clinical interpretability, thereby facilitating reproducibility and prospective applicability at the bedside.
Results: The Eco during labour dataset (n = 124) showed a mean HPD of 3.65 ± 1.29 and a mean AoP of 124.35 ± 16.18°. The Eco in Birth Room (n = 10) showed a mean HPD of 4.33 ± 1.48 and a mean AoP of 112.70 ± 19.68°. Logistic regression determined that increasing HPD was related with a significantly reduced probability of vaginal delivery and a higher probability of caesarean section (β = -4.71, SE = 1.17, p < 0.001). The AoP had a significant inverse correlation with caesarean delivery (β = -0.25, SE = 0.10, p < 0.001). Each one-degree increase in AoP was related with a 23% reduction in the odds of caesarean section (OR = 0.77, 95% CI 0.66-1.00), demonstrating its importance as a reliable indicator of labour progression.
Conclusion: Ultrasound-derived metrics, notably HPD, and AoP provide objective, quantitative indicators of foetal descent that surpass traditional clinical examination. AI-assisted ultrasound analysis may provide continuous, reproducible monitoring of labour dynamics, helping to individualised obstetric care.