Multi-centre deep learning for placenta segmentation in obstetric ultrasound with multi-observer and cross-country generalization

Lisbeth Anita Andreasen, Aasa Feragen, Anders Nymark Christensen, Jonathan Kistrup Thybo, Morten Bo S Svendsen, Kilian Zepf, Karim Lekadir, Martin Grønnebæk Tolsgaard, Lisbeth Anita Andreasen, Aasa Feragen, Anders Nymark Christensen, Jonathan Kistrup Thybo, Morten Bo S Svendsen, Kilian Zepf, Karim Lekadir, Martin Grønnebæk Tolsgaard

Abstract

The placenta is crucial to fetal well-being and it plays a significant role in the pathogenesis of hypertensive pregnancy disorders. Moreover, a timely diagnosis of placenta previa may save lives. Ultrasound is the primary imaging modality in pregnancy, but high-quality imaging depends on the access to equipment and staff, which is not possible in all settings. Convolutional neural networks may help standardize the acquisition of images for fetal diagnostics. Our aim was to develop a deep learning based model for classification and segmentation of the placenta in ultrasound images. We trained a model based on manual annotations of 7,500 ultrasound images to identify and segment the placenta. The model's performance was compared to annotations made by 25 clinicians (experts, trainees, midwives). The overall image classification accuracy was 81%. The average intersection over union score (IoU) reached 0.78. The model's accuracy was lower than experts' and trainees', but it outperformed all clinicians at delineating the placenta, IoU = 0.75 vs 0.69, 0.66, 0.59. The model was cross validated on 100 2nd trimester images from Barcelona, yielding an accuracy of 76%, IoU 0.68. In conclusion, we developed a model for automatic classification and segmentation of the placenta with consistent performance across different patient populations. It may be used for automated detection of placenta previa and enable future deep learning research in placental dysfunction.

Conflict of interest statement

The authors declare no competing interests.

© 2023. The Author(s).

Figures

Figure 1
Figure 1
Confusion matrix for classification of anterior/posterior placentas.
Figure 2
Figure 2
Violin plots of classification accuracy distributions (first row) and distribution of segmentation performance measured by IoU (second row) for different groups across trimesters.
Figure 3
Figure 3
Six annotation examples, including model prediction (red), ground truth (blue) and their overlap (green).

References

    1. Topol EJ. High-performance medicine: The convergence of human and artificial intelligence. Nat. Med. 2019;25:44–56. doi: 10.1038/s41591-018-0300-7.
    1. Sobhaninia Z, et al. Fetal ultrasound image segmentation for measuring biometric parameters using multi-task deep learning. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. 2019;2019:6545–6548.
    1. Garcia-Canadilla P, Sanchez-Martinez S, Crispi F, Bijnens B. Machine learning in fetal cardiology: What to expect. Fetal Diagn. Ther. 2020;47:363–372. doi: 10.1159/000505021.
    1. Pluym ID, et al. Accuracy of automated three-dimensional ultrasound imaging technique for fetal head biometry. Ultrasound Obstet. Gynecol. 2021;57:798–803. doi: 10.1002/uog.22171.
    1. Burgos-Artizzu XP, et al. Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes. Sci. Rep. 2020;10:10200. doi: 10.1038/s41598-020-67076-5.
    1. Hafner E, et al. Comparison between three-dimensional placental volume at 12 weeks and uterine artery impedance/notching at 22 weeks in screening for pregnancy-induced hypertension, pre-eclampsia and fetal growth restriction in a low-risk population. Ultrasound Obstet. Gynecol. 2006;27:652–657. doi: 10.1002/uog.2641.
    1. Collins SL, Stevenson GN, Noble JA, Impey L. Rapid calculation of standardized placental volume at 11 to 13 weeks and the prediction of small for gestational age babies. Ultrasound Med. Biol. 2013;39:253–260. doi: 10.1016/j.ultrasmedbio.2012.09.003.
    1. Yang X, et al. Towards automated semantic segmentation in prenatal volumetric ultrasound. IEEE Trans. Med. Imaging. 2019;38:180–193. doi: 10.1109/TMI.2018.2858779.
    1. Looney P, et al. Fully automated, real-time 3D ultrasound segmentation to estimate first trimester placental volume using deep learning. JCI Insight. 2018;3:120178. doi: 10.1172/jci.insight.120178.
    1. Looney P, et al. Fully automated 3-D ultrasound segmentation of the placenta, amniotic fluid, and fetus for early pregnancy assessment. IEEE Trans. Ultrason. Ferroelectr. Freq. Control. 2021;68:2038–2047. doi: 10.1109/TUFFC.2021.3052143.
    1. Zimmer VA, et al. et al. Towards whole placenta segmentation at late gestation using multi-view ultrasound images. In: Shen D, et al., editors. Medical Image Computing and Computer Assisted Intervention—MICCAI 2019. Springer International Publishing; 2019. pp. 628–636.
    1. Hu R, Singla R, Yan R, Mayer C, Rohling RN. Automated placenta segmentation with a convolutional neural network weighted by acoustic shadow detection. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. 2019;2019:6718–6723.
    1. Zimmer VA. Medical Ultrasound, and Preterm Perinatal and Paediatric Image Analysis. Springer; 2020. A multi-task approach using positional information for ultrasound placenta segmentation; pp. 264–73.
    1. Allen VM, Joseph KS, Murphy KE, Magee LA, Ohlsson A. The effect of hypertensive disorders in pregnancy on small for gestational age and stillbirth: A population based study. BMC Pregnancy Childbirth. 2004;4:17. doi: 10.1186/1471-2393-4-17.
    1. Salafia CM, Charles AK, Maas EM. Placenta and fetal growth restriction. Clin. Obstet. Gynecol. 2006;49:236–256. doi: 10.1097/00003081-200606000-00007.
    1. Malone SL, Haj Yahya R, Kane SC. Reviewing accuracy of first trimester screening for preeclampsia using maternal factors and biomarkers. Int. J. Womens Health. 2022;19(14):1371–1384. doi: 10.2147/IJWH.S283239.
    1. Laganà AS, Giordano D, Loddo S, Zoccali G, Vitale SG, Santamaria A, Buemi M, D'Anna R. Decreased Endothelial Progenitor Cells (EPCs) and increased Natural Killer (NK) cells in peripheral blood as possible early markers of preeclampsia: A case-control analysis. Arch. Gynecol. Obstet. 2017;295(4):867–872. doi: 10.1007/s00404-017-4296-x.
    1. King VJ, Bennet L, Stone PR, Clark A, Gunn AJ, Dhillon SK. Fetal growth restriction and stillbirth: Biomarkers for identifying at risk fetuses. Front. Physiol. 2022;13:959750. doi: 10.3389/fphys.2022.959750.
    1. Telkar N, Stewart GL, Pewarchuk ME, Cohn DE, Robinson WP, Lam WL. Small non-coding RNAs in the human placenta: Regulatory roles and clinical utility. Front. Genet. 2022;13:868598. doi: 10.3389/fgene.2022.868598.
    1. Chiofalo B, Laganà AS, Vaiarelli A, La Rosa VL, Rossetti D, Palmara V, Valenti G, Rapisarda AMC, Granese R, Sapia F, Triolo O, Vitale SG. Do miRNAs play a role in fetal growth restriction? A fresh look to a busy corner. Biomed. Res. Int. 2017;2017:6073167. doi: 10.1155/2017/6073167.
    1. Amodeo S, Cavoretto PI, Seidenari A, Paci G, Germano C, Monari F, Donno V, Giambanco L, Avagliano L, Di Martino D, Fusé F, Masturzo B, Chiantera V, Facchinetti F, Ferrazzi E, Candiani M, Bulfamante G, Farina A. Second trimester uterine arteries pulsatility index is a function of placental pathology and provides insights on stillbirth aetiology: A multicenter matched case-control study. Placenta. 2022;121:7–13. doi: 10.1016/j.placenta.2022.02.021.
    1. Plasencia W, Maiz N, Bonino S, Kaihura C. Uterine artery Doppler at 11 + 0 to 13 + 6 weeks in the prediction of pre-eclampsia. Ultrasound Obstet. Gynecol. 2007;30(5):742–9. doi: 10.1002/uog.5157.
    1. Farina A. Systematic review on first trimester three-dimensional placental volumetry predicting small for gestational age infants. Prenat. Diagn. 2016;36:135–141. doi: 10.1002/pd.4754.
    1. Schwartz N, Wang E, Parry S. Two-dimensional sonographic placental measurements in the prediction of small-for-gestational-age infants. Ultrasound Obstet. Gynecol. 2012;40:674–679. doi: 10.1002/uog.11136.
    1. Burton GJ, Jauniaux E. Pathophysiology of placental-derived fetal growth restriction. Am. J. Obstet. Gynecol. 2018;218:S745–S761. doi: 10.1016/j.ajog.2017.11.577.
    1. Gupta K, et al. Ultrasound placental image texture analysis using artificial intelligence to predict hypertension in pregnancy. J. Matern. Fetal Neonatal Med. 2021 doi: 10.1080/14767058.2021.1887847.
    1. Wada, K. Labelme. . (2016).
    1. He, K., Gkioxari, G, D., P. & Girschick R. Mask R-CNN. In 2017 IEEE International Conference on Computer Vision (ICCV) 2980–2988. 10.1109/ICCV.2017.322 (2017).
    1. Paszke A. Pytorch: An imperative style, high-performance deep learning library. Adv. Neural Inf. Process. Syst. 2019;2019:8026–8037.
    1. .

Source: PubMed

Sponsors et collaborateurs

Les conditions médicales

Interventions en matière de drogue

3
S'abonner