OC10.04: Estimating fetal gestational age based on ultrasound image characteristics using artificial intelligence

Elizabeth Bradburn, Lok Hin Lee, J. Alison Noble, Aris T. PapageorghiouUltrasound 2020 · DOI

Abstract

Assessment of gestational age (GA) is essential to guide obstetric care. Measurement of CRL is the gold standard for GA estimation, but when women present in late pregnancy, GA estimation is based on biometry. This is inaccurate when fetal size is smaller or larger than expected. Our aim was to assess if GA can be accurately estimated based on automated image appearance analysis, without fetal biometry. Quality assessed ultrasound images of head circumference (HC) and abdominal circumference (AC) were used from a multicentre study database of women with known GA (CRL performed < 14 weeks) who underwent serial ultrasound examinations from 15 to 40 weeks. We split data randomly into 67.5% training, 7.5% validation and 25% testing sets. We developed and tested an artificial intelligence (AI) algorithm using a regression neural network to estimate the GA, based on image characteristics only without measurement information. In total 24,403 ultrasound images were used. Automatic GA estimation was most accurate in the early second trimester and variation increased as GA increased. The HC performed best: at 16 weeks the estimated GA with variation around the mean (95% Limits of Agreement) was +/- 7.8 days, while at 32 weeks it was +/- 17.2 days. For the AC the respective values were +/- 15.0 days and +/- 18.8 days. This compares with previously published formula which estimates GA from HC and femur length (FL) measurements +/- 6.5 days and +/- 15.4 days, respectively. These early results of AI-based GA estimation compare favourably with estimation based on biometry. As our method is based only on image characteristics, the theoretical advantage is that it is less prone to GA estimation error when there are fetal growth aberrations. This has the potential to improve GA estimation in settings where women first present late in pregnancy.
BibTeX
@inproceedings{bradburn2020oc,
  author = {Elizabeth Bradburn and Lok Hin Lee and J. Alison Noble and Aris T. Papageorghiou},
  title = {{OC10.04: Estimating fetal gestational age based on ultrasound image characteristics using artificial intelligence}},
  booktitle = {Ultrasound},
  year = {2020},
  doi = {10.1002/uog.22268},
}

Loading related content...

Figure