Moderate evidence
More
AI bone age tool matched radiologists in a study of 1,516 children's X-rays
Study title: Beyond chronological age: uncovering disease-dependent variability in AI-driven pediatric bone age assessment
- Evidence type
- Observational study
- Sample
- n = 1,516
- Sources
- 1 sourcePeer-reviewed
- Published
- 2 min read
- Truvace record
- TRV-2026-2966cite or check it
Study title: Beyond chronological age: uncovering disease-dependent variability in AI-driven pediatric bone age assessment
A retrospective study in Insights into Imaging compared the commercial BoneAge AI tool with routine bone age reports by board-certified pediatric radiologists on 1,516 hand radiographs. The children had a mean age of 10.84 years, and 50.6% were boys.
Read the study abstract
Objectives To assess performance and agreement of a commercial artificial intelligence (AI) tool relative to routine pediatric radiologists' bone age assessments in a large pediatric cohort, with focus on different groups of growth disorders. Materials and methods One thousand five hundred sixteen anterior-posterior hand radiographs of pediatric patients (50.6% boys, mean age 10.84 ± 3.66 years) were retrospectively analyzed using the AI tool BoneAge. AI results were compared with board-certified pediatric radio…

The gain
In a retrospective study of 1,516 children's hand X-rays, the BoneAge AI tool differed from pediatric radiologists' bone age readings by a mean of 6.4 months, with correlation of 0.980 and intraclass correlation above 0.9.
In brief
The mean absolute difference between AI and radiologists was 6.4 months, and the two gave exactly the same result in 45.1% of cases. Correlation across the whole cohort was 0.980, and intraclass correlation exceeded 0.9.
Agreement varied by the reason for imaging. Orthopedic cases matched best, with a mean difference of 4.1 months, while genetic/neuroendocrine cases differed by 6.6 months and oncologic cases by 6.7 months. The authors say radiologists should keep disease-specific differences in mind when using automated tools.
Main points
- The study analysed 1,516 anterior-posterior hand radiographs of children (mean age 10.84 ± 3.66 years; 50.6% boys).
- AI output was compared with pediatric radiologists' reports using Bland-Altman plots, Spearman correlations and intraclass correlation coefficients.
- Mean absolute difference was 6.4 ± 7.0 months, with complete concordance in 45.1% of cases.
- Orthopedic cases (n = 152) had the smallest difference, 4.1 ± 5.6 months.
- Genetic/neuroendocrine cases (n = 1,179) differed by 6.6 ± 7.2 months and oncologic cases (n = 185) by 6.7 ± 6.3 months.
- Spearman correlations were 0.974 for genetic/neuroendocrine, 0.952 for orthopedic and 0.988 for oncologic indications.
The full story
The researchers retrospectively analysed routine hand radiographs, setting the BoneAge software's estimates against reports already written by board-certified pediatric radiologists. This makes the comparison a real-world one rather than a test on a curated benchmark dataset.
Across all 1,516 children, agreement was high: Spearman correlation was 0.980 and intraclass correlation exceeded 0.9 for the whole cohort and each indication group. The average gap was 6.4 months, and the AI and radiologists gave identical results in 45.1% of cases.
Subgroup results differed. Orthopedic cases had the smallest gap (4.1 months), while the genetic/neuroendocrine group, the largest at 1,179 children, and the oncologic group had gaps near 6.6 and 6.7 months. The authors say this is the first study of this tool to consider the indication for imaging, and that radiologists should account for disease-specific differences when using automated tools.
Sources
1 on file- Insights into Imagingdoi:10.1186/s13244-026-02477-0
Talk it over
The BoneAge AI tool differed from pediatric radiologists by a mean of 6.4 months and gave identical results in 45.1% of cases. How would you weigh those two figures when judging whether it is good enough?
AI bone age tool matched radiologists in a study of 1,516 children's X-rayshttps://truvace.com/item/beyond-chronological-age-uncovering-disease-dependent-variability-in-ai-driven-p
The BoneAge AI tool differed from pediatric radiologists by a mean of 6.4 months and gave identical results in 45.1% of cases. How would you weigh those two figures when judging whether it is good enough?
Answer in the threadThe researchers compared the tool with radiologists' routine reports on 1,516 hand X-rays, not with a separate gold standard. What does that comparison tell us, and which question about accuracy does it leave open?
Answer in the threadOrthopedic cases (n = 152) differed by 4.1 months, while genetic/neuroendocrine cases (n = 1,179) differed by 6.6 months and oncologic cases (n = 185) by 6.7 months. Which children might benefit least, and how would you find out?
Answer in the thread
If you have more time
If you were a radiologist using BoneAge AI, how would the indication for imaging change how closely you check its estimate?
Answer in the threadWhat follow-up study, with which groups of children, would show whether a 6.4-month mean difference matters for treatment decisions?
Answer in the thread
ace
What readers say