Related Experiment Video
Updated: Jan 11, 2026
Assessment of Bone Fracture Healing Using Micro-Computed Tomography
Published on: December 9, 2022
Multivendor comparison study of artificial intelligence software for automated fracture detection in paediatric
Irmhild Altmann-Schneider1,2, Léa S Geiger3, Christian J Kellenberger4
1Department of Diagnostic Imaging, University Children's Hospital Zurich, Lenggstrasse 30, Zurich, 8008, Switzerland. irmhild.altmann@kispi.uzh.ch.
Background:
Evidence for artificial intelligence (AI)-assisted paediatric fracture detection is limited. External validation and comparison of AI software are required before reliable use in clinical practice.
Objective:
To evaluate and compare the performance of three commercially available AI software for detecting posttraumatic findings in paediatric patients.
Materials And Methods:
This retrospective study assessed three AI software using radiographs of children aged 2-17 years who presented to the emergency department after trauma. Radiographs of the lower leg, forearm, and elbow were included between January 2014 and January 2024 (lower leg), March 2022 and January 2024 (forearm), and July 2019 and January 2024 (elbow). Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for fractures, effusions, and dislocations.
Results:
A total of 3,013 patients with 3,414 radiographs were included: 1,074 lower leg (mean age 6.6 years), 1,142 forearm (7.4 years), and 1,198 elbow (7.5 years). All AI tools demonstrated high performance for lower leg and forearm radiographs, with sensitivity of 88.4-94.7%, specificity 93.6-99.2%, PPV 94.5-99.2%, and NPV 91.6-95.6%. In contrast, performance for elbow radiographs was reduced (sensitivity 72.8-91.6%, specificity 80.3-98.7%), with the lowest PPV of 86.1% and NPV of 79.5%. Sensitivity was notably reduced for specific paediatric fracture types, elbow effusions (posterior fat pad sign 40.6-82.3%), and dislocations (54.2-93.8%), with significant differences between AI software.
Conclusions:
AI tools show promise for paediatric fracture detection, particularly in lower leg and forearm radiographs. Awareness of their limitations is essential for safe clinical use.
