Artificial Intelligence in Bone Age Assessment of Pediatric Hand-Wrist X-Rays: Systematic Review and Meta-Analysis of Studies from 2019–2024

Abstract

Objective: To explore the existing literature on the use of artificial intelligence (AI) and machine learning (ML) in bone age assessment (BAA). Methodology: This study was conducted in accordance with the PRISMA guidelines. Three databases (PubMed, Scopus, and Co-chrane) were screened for studies published from 2019–2024. Two reviewers independently selected the studies. The modified Kitchenham-Charters’ checklist was used to critically appraise the studies. Only studies reporting the mean absolute error (MAE) were included in the analysis. Data regarding the study character-istics, subject characteristics, ML technique, model ground truth, and the performance of the studies were extracted. Results: The review included 33 studies, mostly from East Asia. Most studies used in-house datasets consisting of hand radiographs of their respective local population. Convolutional neural networks are the most popular AI algorithms used. Most studies used radiologist-annotated BA as ground truth rather than the chronological age. The meta-analysis revealed a weighted MAE of 7.54 months, an improvement compared to the previous study. Conclusion: AI and ML models continue to demonstrate rapid ad-vancements in their application for BAA. This study described the current trends in ML research and explored ongoing obstacles in BAA, as well as the prospective role of AI. While promising, further research is still required to address current limitations, such as validity issues. Subsequent studies should also be conducted with rigorous methodology and thorough reporting.