Popular Boards

David Oelen, Pascal Kaiser, Thomas Baumann | Ultraschall in der Medizin - European Journal of Ultrasound | (2020)

Abstract

The accuracy of physicians in their daily routine is inferior to deep learning-based algorithms for determining angles in ultrasound of the newborn hip. Similar methods could be used to support physicians.

Tags

Sample Definition And Size

The study used 303,306 ultrasound images of newborn hips collected between 2009 and 2016 during screening consultations. Trained physicians labeled every second image with alpha and beta angles during routine consultations. A random subset of images was labeled under laboratory conditions with time and precision to serve as ground truth. The analysis focused on the alpha angle. ([thieme-connect.de](https://www.thieme-connect.de/products/all/doi/10.1055/a-1177-0480?utm_source=openai))

Study Type

This was an observational study comparing human performance (trained physicians) to automated methods (deep learning-based convolutional neural networks) in measuring angles in ultrasound images. Three CNN-based methods were implemented: (1) direct angle prediction, (2) landmark detection, and (3) detection of baseline and roof lines for angle calculation. ([thieme-connect.de](https://www.thieme-connect.de/products/all/doi/10.1055/a-1177-0480?utm_source=openai))

Conflicts Of Interest

No conflicts of interest are declared in the accessible abstract or metadata. ([thieme-connect.de](https://www.thieme-connect.de/products/all/doi/10.1055/a-1177-0480?utm_source=openai))

Results Summary

The root mean squared error (RMSE) between physicians’ measurements and ground truth for the alpha angle was 7.1°. The best-performing CNN method (landmark detection) achieved an RMSE of 3.9° for the alpha angle. ([thieme-connect.de](https://www.thieme-connect.de/products/all/doi/10.1055/a-1177-0480?utm_source=openai))

Referenced In

🍼 Why reading baby hip ultrasounds is harder than you think—and how AI is stepping in to help.

Ever had an ultrasound? Now imagine trying to get a perfect picture of a squirming infant's hip joint while they're crying and moving. That's the reality of screening for developmental dysplasia of the hip (DDH), a condition affecting 1–3% of babies that can cause lifelong disability if missed.

The problem? 🎯 Too few experts, too many babies, and results that vary wildly depending on who's holding the probe.

A new systematic review & meta-analysis by Azmi et al., 2026, just published in Pediatric Radiology, brings together the strongest evidence yet on AI-assisted infant hip ultrasound—and the results are promising:

🔍 Key Findings from Azmi et al. (2026):

  • High diagnostic accuracy: Pooled sensitivity 92% (95% CI 86–95%) and specificity 96% (95% CI 91–98%) across 9 studies with 6,351 hips

    • Compiled 29 studies but analysed only 9 who provided 2×2 data using bivariate random-effects model.

  • Fast to learn: Operators need only ~1–2 hours of training to use AI-assisted systems

  • Quicker scans: Acquisition times drop by 20–50%

🤩 Compared to other reviews like Bhavsar et al., 2025 & Kamath et al., 2026, this meta-analysis analysed raw imaging data rather than final statistical predictions, making this a promising census for DDH. The authors then close the discussion by calling for multi-centre trials - testing that spans across different hospitals and environments.

What's your take? Is AI ready to assist or even replace the expert eye?

Fun Fact🤓: The "Graf method" requires precise positioning and angle measurements that take years to master. Miss the angle by a few degrees, and you miss the diagnosis or even send healthy babies for unnecessary treatments.

  • A 2022 paper by Oelen et al., 2022 claims that AI precision outperforms trained physicians, with an error of 3.9° vs. physician error of 7.1° for alpha angles

If you'd like a comprehensive read on the current trends in paediatric orthopaedic disorders, check out this review on recent developments: Current Trends and Future Directions in the Diagnosis and Management of Pediatric Orthopedic Disorders

Image credit: kidshealth.org

1