Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Mar 9;6(1):36.
doi: 10.1038/s41746-023-00774-2.

Machine learning for accurate estimation of fetal gestational age based on ultrasound images

Lok Hin Lee #  1 , Elizabeth Bradburn #  2 , Rachel Craik  2 , Mohammad Yaqub  3 , Shane A Norris  4 , Leila Cheikh Ismail  5 , Eric O Ohuma  2   6 , Fernando C Barros  7   8 , Ann Lambert  2 , Maria Carvalho  9 , Yasmin A Jaffer  10 , Michael Gravett  11 , Manorama Purwar  12 , Qingqing Wu  13 , Enrico Bertino  14 , Shama Munim  15 , Aung Myat Min  16 , Zulfiqar Bhutta  15   17 , Jose Villar  2   18 , Stephen H Kennedy  2   18 , J Alison Noble #  1 , Aris T Papageorghiou #  19   20
Affiliations

Machine learning for accurate estimation of fetal gestational age based on ultrasound images

Lok Hin Lee et al. NPJ Digit Med. .

Abstract

Accurate estimation of gestational age is an essential component of good obstetric care and informs clinical decision-making throughout pregnancy. As the date of the last menstrual period is often unknown or uncertain, ultrasound measurement of fetal size is currently the best method for estimating gestational age. The calculation assumes an average fetal size at each gestational age. The method is accurate in the first trimester, but less so in the second and third trimesters as growth deviates from the average and variation in fetal size increases. Consequently, fetal ultrasound late in pregnancy has a wide margin of error of at least ±2 weeks' gestation. Here, we utilise state-of-the-art machine learning methods to estimate gestational age using only image analysis of standard ultrasound planes, without any measurement information. The machine learning model is based on ultrasound images from two independent datasets: one for training and internal validation, and another for external validation. During validation, the model was blinded to the ground truth of gestational age (based on a reliable last menstrual period date and confirmatory first-trimester fetal crown rump length). We show that this approach compensates for increases in size variation and is even accurate in cases of intrauterine growth restriction. Our best machine-learning based model estimates gestational age with a mean absolute error of 3.0 (95% CI, 2.9-3.2) and 4.3 (95% CI, 4.1-4.5) days in the second and third trimesters, respectively, which outperforms current ultrasound-based clinical biometry at these gestational ages. Our method for dating the pregnancy in the second and third trimesters is, therefore, more accurate than published methods.

PubMed Disclaimer

Conflict of interest statement

J.A.N. and A.T.P. are Senior Scientific Advisors of Intelligent Ultrasound Ltd. All other authors declare that they have no competing interests or other interests that might be perceived to influence the results and/or discussion reported in this paper.

Figures

Fig. 1
Fig. 1. Performance per plane and MutliPlane on the INTERGROWTH-21st internal validation set.
Composite figure showing scatter plot of the performance of the proposed single standard plane and MultiPlane models on the hold-out INTERGROWTH-21st test set with histograms showing the spread of planes across gestational ages. HCP Head Circumference Plane, ACP Abdominal Circumference Plane, FLP Femur Length Plane.
Fig. 2
Fig. 2. Performance per plane and MultiPlane on the INTERBIO-21st external validation set.
Composite figure showing scatter plot of the performance of the proposed single standard plane and MultiPlane models on the INTERBIO-21st test set with histograms showing the spread of planes across gestational ages. HCP Head Circumference Plane, ACP Abdominal Circumference Plane, FLP Femur Length Plane.
Fig. 3
Fig. 3. Modified Bland-Altman plots showing per plane performance on the INTERGROWTH-21st internal validation set.
Modified Bland-Altman plots of the performance of the gestational age (GA) estimation models on the INTERGROWTH-21st dataset. The striations visible on the data points for the single standard plane models are due to the CORAL binning process employed during training, which is removed for the final MultiPlane regression loss. Due to the variation in the number of data points per figure, the transparency of each scatter point was normalised so that the overall appearance is normalised. HCP Head Circumference Plane; GA Gestational Age; CORAL Consistent Ordinal RAnk Logits; GT Ground Truth; SD Standard Deviation; ACP Abdominal Circumference Plane; FLP Femur Length Plane.
Fig. 4
Fig. 4. MultiPlane performance in small and large for gestational age fetuses.
Gestational age (GA) estimation results in pregnancies resulting in small for gestational age (SGA) or large for gestational age (LGA) newborns, defined by a birth weight >90th centile (LGA) or <10th centile (SGA) according to the INTERGROWTH-21st international standard.
Fig. 5
Fig. 5. Training process for single plane estimation of gestational age.
Schematic of training process for single ultrasound standard plane-based gestational age (GA) estimation using Consistent Rank Logit Loss. HC Head Circumference; GA Gestational Age; AC Abdominal Circumference; FL Femur Length.
Fig. 6
Fig. 6. Training process for MultiPlane model for gestational age estimation.
Schematic of training process for multiple ultrasound standard plane-based gestational age (GA) estimation. Pre-trained models are based on single standard plane images trained with Consistent Rank Logit Loss. The final multiple standard plane model is trained with L1-loss. HC Head Circumference; AC Abdominal Circumference; FL Femur Length; GA Gestational Age.
See this image and copyright information in PMC

References

    1. March of Dimes, P., Save the Children. WHO. Born Too Soon The Global Action Report on Preterm Birth (eds Howson, C. P., Kinney, M. V. & Lawn, J. E.) (World Health Organization, Geneva, 2012).
    1. Perin J, et al. Global, regional, and national causes of under-5 mortality in 2000–19: an updated systematic analysis with implications for the Sustainable Development Goals. Lancet Child Adolesc. Health. 2022;6:106–115. doi: 10.1016/S2352-4642(21)00311-4. - DOI - PMC - PubMed
    1. Brighton A, D’Arcy R, Kirtley S, Kennedy S. Perceptions of prenatal and obstetric care in Sub-Saharan Africa. Int. J. Gynaecol. Obstet. 2013;120:224–227. doi: 10.1016/j.ijgo.2012.09.017. - DOI - PubMed
    1. W. H. O. Alliance for Maternal Newborn Health Improvement Late Pregnancy Dating Study Group. Performance of late pregnancy biometry for gestational age dating in low-income and middle-income countries: a prospective, multicountry, population-based cohort study from the WHO Alliance for Maternal and Newborn Health Improvement (AMANHI) Study Group. Lancet Glob. Health. 2020;8:e545–e554. doi: 10.1016/S2214-109X(20)30034-6. - DOI - PMC - PubMed
    1. Savitz DA, et al. Comparison of pregnancy dating by last menstrual period, ultrasound scanning, and their combination. Am. J. Obstet. Gynecol. 2002;187:1660–1666. doi: 10.1067/mob.2002.127601. - DOI - PubMed