Low‐Field Magnetic Resonance Imaging of the Late Gestation Cervix and Birth Outcome Correlation: A Prospective Cohort Study

ABSTRACT Objective To use low‐field MRI to produce reconstructions and 3D models of the cervix and to automate measurements for correlation with demographics and birth outcomes. Design Prospective cohort study. Setting KCL Advanced Imaging Centre, St Thomas's Hospital. Population Late gestation (36‐41w) women attempting their first vaginal birth, recruited to the MiBirth study ( n = 97). Methods Reconstructed images were produced from 2D T2‐weighted Turbo‐Spin‐Echo 2D sequences acquired with a 0.55 T Freemax MRI scanner. Segmentations and anatomical landmarks were automated using an in‐house 3D deep learning segmentation network, from which cervical 2D measurements and 3D volumes were generated. Main Outcome Measures Quality of reconstructed images and segmentations. Inter‐rater variability for cervical biometry. Correlation between cervical measurements, maternal demographics and birth outcomes. Results Successful reconstructions were obtained for 92.9%; 84.9% were good quality. Excellent or good quality segmentations were obtained for all successful reconstructions ( n = 99). Inter‐rater variability between automated and manual biometry was excellent or good for cervical measurements. Total cervical and stroma volumes significantly increased with cervical length ( p < 0.01). Os diameters and utero‐cervical angle significantly decreased as cervical length increased ( p < 0.001). Cervical stroma volume increased with maternal age ( p = 0.02). Controlling for maternal age, an increased cervical volume was associated with an increased risk of caesarean section (OR 1.09, p = 0.04). Conclusions This is a novel, accurate automated system to assess MRI late gestation cervical biometry and volumetry. We have shown that the late gestation cervical phenotype may influence birth outcomes and provided a new mechanism for increased risk of caesarean with maternal age.