Bone age estimation from chest radiographs using deep neural networks: a proof-of-concept study

Area of Science:

• Pediatric radiology
• Artificial intelligence in medicine
• Skeletal maturity assessment

Background:

• Bone age (BA) assessment is crucial for pediatric growth evaluation but relies on hand radiographs, which are not always practical.
• Current methods require specialized equipment and expert interpretation, limiting routine use.

Purpose of the Study:

• To develop a deep neural network model for estimating bone age from standard pediatric chest radiographs.
• To assess the feasibility of using AI for opportunistic bone age estimation.

Main Methods:

• Retrospective analysis of 101 children (3-15 years) with both chest and hand radiographs.
• Exclusion of patients with skeletal dysplasia or chest wall abnormalities.
• Development and fine-tuning of deep neural networks using sex-considering and sex-non-considering models to estimate BA from chest X-rays, validated against the Tanner-Whitehouse 2 radius-ulna-short bones (TW2-RUS) method.

Main Results:

• The deep learning models demonstrated good concordance with reference bone age standards.
• Sex-considering models achieved higher accuracy, with intraclass correlation coefficients (ICCs) up to 0.87 and root mean squared errors (RMSEs) as low as 1.30.
• Sex-non-considering models showed comparable results with ICCs up to 0.81 and RMSEs of 1.52.

Conclusions:

• Deep neural networks can accurately estimate bone age from pediatric chest radiographs.
• This AI-driven approach presents a viable, opportunistic alternative to dedicated hand radiographs for skeletal maturity assessment.
• Further validation could integrate this method into routine pediatric imaging workflows.