Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
175 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
42 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Whole-examination AI estimation of fetal biometrics from 20-week ultrasound scans (2401.01201v1)

Published 2 Jan 2024 in cs.CV and cs.LG

Abstract: The current approach to fetal anomaly screening is based on biometric measurements derived from individually selected ultrasound images. In this paper, we introduce a paradigm shift that attains human-level performance in biometric measurement by aggregating automatically extracted biometrics from every frame across an entire scan, with no need for operator intervention. We use a convolutional neural network to classify each frame of an ultrasound video recording. We then measure fetal biometrics in every frame where appropriate anatomy is visible. We use a Bayesian method to estimate the true value of each biometric from a large number of measurements and probabilistically reject outliers. We performed a retrospective experiment on 1457 recordings (comprising 48 million frames) of 20-week ultrasound scans, estimated fetal biometrics in those scans and compared our estimates to the measurements sonographers took during the scan. Our method achieves human-level performance in estimating fetal biometrics and estimates well-calibrated credible intervals in which the true biometric value is expected to lie.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (19)
  1. L. J. Salomon, Z. Alfirevic, F. D. S. Costa, R. L. Deter, F. Figueras, T. Ghi, P. Glanc, A. Khalil, W. Lee, R. Napolitano, A. Papageorghiou, A. Sotiradis, J. Stirnemann, A. Toi, and G. Yeo, “Isuog practice guidelines: ultrasound assessment of fetal biometry and growth,” Ultrasound in Obstetrics and Gynecology, vol. 53, pp. 715–723, 6 2019.
  2. N. H. S. England, “Fetal anomaly screening programme handbook,” 2021.
  3. L. J. Salomon, Z. Alfirevic, V. Berghella, C. M. Bilardo, G. E. Chalouhi, F. D. S. Costa, E. Hernandez-Andrade, G. Malinger, H. Munoz, D. Paladini, F. Prefumo, A. Sotiriadis, A. Toi, and W. Lee, “Isuog practice guidelines (updated): performance of the routine mid-trimester fetal ultrasound scan,” Ultrasound in Obstetrics and Gynecology, vol. 59, pp. 840–856, 6 2022.
  4. L. Drukker, R. Droste, P. Chatelain, J. A. Noble, and A. T. Papageorghiou, “Expected-value bias in routine third-trimester growth scans,” Ultrasound in Obstetrics and Gynecology, vol. 55, pp. 375–382, 3 2020.
  5. I. Sarris, C. Ioannou, P. Chamberlain, E. Ohuma, F. Roseman, L. Hoch, D. G. Altman, and A. T. Papageorghiou, “Intra- and interobserver variability in fetal ultrasound measurements,” Ultrasound in Obstetrics and Gynecology, vol. 39, pp. 266–273, 2012.
  6. C. F. Baumgartner, K. Kamnitsas, J. Matthew, T. P. Fletcher, S. Smith, L. M. Koch, B. Kainz, and D. Rueckert, “Sononet: Real-time detection and localisation of fetal standard scan planes in freehand ultrasound,” IEEE Transactions on Medical Imaging, vol. 36, 2017.
  7. H. Chen, Q. Dou, D. Ni, J. Z. Cheng, J. Qin, S. Li, and P. A. Heng, “Automatic fetal ultrasound standard plane detection using knowledge transferred recurrent neural networks,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9349, pp. 507–514, 2015.
  8. X. P. Burgos-Artizzu, D. Coronado-Gutierrez, B. Valenzuela-Alcaraz, E. Bonet-Carne, E. Eixarch, F. Crispi, and E. Gratacos, “Evaluation of deep convolutional neural networks for automatic classification of common maternal fetal ultrasound planes,” Scientific Reports 2020 10:1, vol. 10, pp. 1–12, 6 2020.
  9. M. Sinclair, C. F. Baumgartner, J. Matthew, W. Bai, J. C. Martinez, Y. Li, S. Smith, C. L. Knight, B. Kainz, J. Hajnal, A. P. King, and D. Rueckert, “Human-level performance on automatic head biometrics in fetal ultrasound using fully convolutional neural networks,” Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS, vol. 2018-July, pp. 714–717, 10 2018.
  10. M. Yaqub, N. Sleep, S. Syme, Z. Chen, H. Ryou, S. Walton, J. A. Noble, and A. T. Papageorghiou, “491 scannav ® audit: an ai-powered screening assistant for fetal anatomical ultrasound,” American Journal of Obstetrics and Gynecology, vol. 224, p. S312, 2 2021.
  11. S. Plotka, T. Wlodarczyk, A. Klasa, M. Lipa, A. Sitek, and T. Trzcinski, “Fetalnet: Multi-task deep learning framework for fetal ultrasound biometric measurements,” Communications in Computer and Information Science, vol. 1517 CCIS, pp. 257–265, 2021.
  12. C. Lee, A. Willis, C. Chen, M. Sieniek, A. Watters, B. Stetson, A. Uddin, J. Wong, R. Pilgrim, K. Chou, D. Tse, S. Shetty, and R. G. Gomes, “Development of a machine learning model for sonographic assessment of gestational age,” JAMA Network Open, vol. 6, pp. e2248685–e2248685, 1 2023.
  13. J. Matthew, E. Skelton, T. G. Day, V. A. Zimmer, A. Gomez, G. Wheeler, N. Toussaint, T. Liu, S. Budd, K. Lloyd, R. Wright, S. Deng, N. Ghavami, M. Sinclair, Q. Meng, B. Kainz, J. A. Schnabel, D. Rueckert, R. Razavi, J. Simpson, and J. Hajnal, “Exploring a new paradigm for the fetal anomaly ultrasound scan: Artificial intelligence in real time,” Prenatal Diagnosis, vol. 42, pp. 49–59, 1 2022.
  14. R. Smith, “An overview of the tesseract ocr engine,” Proceedings of the International Conference on Document Analysis and Recognition, ICDAR, vol. 2, pp. 629–633, 2007.
  15. Voluson E8/E8 Expert Basic User Manual. GE Healthcare, 2012.
  16. B. 3rd Trimester Special Interest Group, “Professional guidance for fetal growth scans performed after 23 weeks of gestation,” 2022.
  17. M. R. Chavez, C. V. Ananth, J. C. Smulian, S. Lashley, E. V. Kontopoulos, and A. M. Vintzileos, “Fetal transcerebellar diameter nomogram in singleton gestations with special emphasis in the third trimester: A comparison with previously published nomograms,” American Journal of Obstetrics and Gynecology, vol. 189, pp. 1021–1025, 10 2003.
  18. “Labelbox | data-centric ai platform for building intelligent applications.”
  19. S. Constantine, A. Kiermeier, and P. Anderson, “The normal fetal cephalic index in the second and third trimesters of pregnancy,” Ultrasound Quarterly, vol. 36, pp. 255–262, 2020.
Citations (1)
View on Semantic Scholar

Summary

  • The paper introduces a fully automated AI system that uses CNN and Bayesian methods to estimate fetal biometrics from every frame of a 20-week ultrasound scan.
  • The methodology minimizes human bias by aggregating measurements across frames, achieving performance comparable to expert sonographers.
  • The system’s potential to reduce scan time and cognitive load highlights its promise for clinical adoption and enhanced prenatal diagnostics.

Introduction

AI is transforming medical imaging and diagnostics. A paper explores an AI system's ability to estimate fetal biometrics from ultrasound scans. Conventionally, fetal biometrics are manually measured by sonographers from individually selected images, which can introduce biases and variability. The presented system uses a Convolutional Neural Network (CNN) and a Bayesian method to estimate biometrics from every frame of a 20-week ultrasound, aiming to provide human-level performance while reducing operator error and cognitive load.

AI-Driven Biometric Estimation

The system uses a CNN to analyze each frame of an ultrasound video, measuring visible fetal anatomy. A Bayesian method computes the true value of each biometric, aggregating data from all frames and factoring out outliers. In a retrospective experiment with over 1,457 recordings, the AI-generated estimates rivaled the measurements taken by sonographers during standard scans, suggesting high accuracy and well-calibrated credible intervals.

Method and Dataset

The AI model was trained using the iFIND project dataset with 20-week scans, following UK protocols. It involved different machine resolutions and sonographers manually capturing standard plane images and biometrics. The team removed the sonographer's annotations to avoid bias, designing the AI to automatically classify each frame and measure biometrics in real-time, discarding anatomically implausible values. This automation attempts to minimize human biases and errors, producing a single estimate from the vast data collected.

Outcomes and Potential

The findings show that automated full-scan biometric estimation can potentially yield more accurate and consistent results than manual methods, even outperforming human sonographers in terms of repeatability and consistency. Additional benefits, such as reduced scan time and increased sonographer focus, are anticipated though not quantified in this paper. A patent has been filed related to this method, and the research team, including co-founders of Fraiya Ltd, seeks to commercialize the technology. A prospective trial is underway to determine the system's efficacy in clinical settings and its impact on the detection of fetal abnormalities.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets