VP34.21: Machine learning of the basic obstetric ultrasound examination: validation of the standardised six‐step approach

Alice Self, Q. Chen, Shinjini Bhatnagar, J.a. Noble, Aris T. PapageorghiouUltrasound 2020 · DOI

Abstract

In low-income settings women have limited access to antenatal ultrasound. A combination of low-cost ultrasound devices, simple imaging protocols and intelligent image analysis could enable less skilled operators to perform basic obstetric scans reliably. We assess the validity of the six-step protocol (6SP), designed for sonographer use, for machine learning purposes. Pregnant women at 19 to 36 weeks were enrolled prospectively and the 6SP undertaken by trained sonographers. Retrospective ultrasound video analysis assessed if the first five steps could independently and accurately confirm presentation, viability, multiple pregnancy, placentation and amniotic fluid. 230 videos were recorded from 1 twin and 79 singleton pregnancies. Placental location could not be confirmed by any step in 11/80 cases. Step 1 could not confirm fetal presentation in 31/77 cases and Step 2 could not confirm viability in 15/77 cases. If all other steps were considered, presentation and viability could be confirmed in 100% and 99%, respectively. There was one case of oligohydramnios, correctly identified by the 6SP. In clinical studies, the 6SP has one step for each objective. We show that meaningful, overlapping clinical information can be obtained from different steps of the 6SP. Just as humans do not interpret single scan sweeps in isolation, this study suggests there may be benefits for automated decision-making algorithms to be based on scanning protocols in their entirety. A way to better visualise the placenta is needed.
BibTeX
@inproceedings{self2020vp,
  author = {Alice Self and Q. Chen and Shinjini Bhatnagar and J.A. Noble and Aris T. Papageorghiou},
  title = {{VP34.21: Machine learning of the basic obstetric ultrasound examination: validation of the standardised six‐step approach}},
  booktitle = {Ultrasound},
  year = {2020},
  doi = {10.1002/uog.22854},
}

Loading related content...

Figure