Construction of a Simplified Bone Age Assessment Model in Chinese Children Aged 3-18 Years

Abstract

Context: Bone age assessment (BAA) is critical for pediatric endocrinology. Traditional methods are complex and time-consuming, and current computer-aided systems have made progress but still lacking in robustness.

Objective: Develop simplified BAA methods to aid pediatricians in quick clinical assessments.

Design: Overall, 5,551 left-hand X-ray images were collected from a cross-sectional survey in 2022- 2023. Bone age (BA) was assessed using the Tanner-Whitehouse 3-China/radius-ulna-short bone (TW3-C-RUS) method. The linear regression models were constructed with BA as the dependent variable and 13 RUS bones' grades as independent variables.

Setting: A cross-sectional survey in Beijing, China.

Participants: 5,551 children aged 3-18 years.

Main outcome measures: Model accuracy was evaluated by R², residuals, and root mean square error (RMSE), and BA with an error margin ≤0.5 years.

Results: When bone grades were consistent, a single or few bones may serve as proxies for BA, such as metacarpals and phalanges reached grade 6, the BA for boys and girls were 13.0-13.5 years and 10.9-11.5 years, respectively. When bone grades were inconsistent, regression models were needed. Comparatively, the three-bone models have advantages and are proposed for clinically simplified BAA. The simplified three-bone model (radius, ulna, and metacarpal I) yielded an R² > 0.94 and an RMSE < 0.5 years. When considering puberty stages, specific bone combination models can further improve assessment accuracy.

Conclusion: These three-bone models could be useful for rapid BAA, with improved accuracy when puberty stages is included. Further validation is warranted to test their robustness across populations and scenarios.

Keywords: Bone age; Children; Clinical simplified model; TW3-C-RUS.