Fetal biometry and amniotic fluid volume assessment end-to-end automation using Deep Learning

Abstract

Fetal biometry (FB) and amniotic fluid volume (AFV) assessments are two crucial yet repetitive tasks of fetal ultrasound screening scans that help detect potential life-threatening conditions, however, they suffer from reproducibility and reliability issues. Advances in deep learning have led to new applications in measurement automation in fetal ultrasound, showcasing human-level performances in several fetal ultrasound tasks. However, most of the studies performed are retrospective “in silico” studies and few include African patients in their dataset. Here we develop and prospectively assess the performance of deep learning models for an end-to-end FB and AFV automation from a newly constructed database of 172 293 de-identified Moroccan fetal ultrasound images in addition to publicly available datasets. They were tested on prospectively acquired video clips from 172 patients forming a consecutive series gathered at four healthcare centers in Morocco. Our results show the 95% limits of agreement between the models and practitioners for the studied measurements were narrower than reported intra and inter-observer variability for human expert sonographers for all the studied parameters. This means that these models could be deployed in clinical conditions, to alleviate time-consuming, repetitive tasks, and to make fetal US more accessible in limited resources environments.