A Comprehensive Review of Artificial Intelligence Methods in Bone Age Assessment

Bone age reflects individual skeletal maturity and is an important factor in the follow-up and monitoring of growth and development in children. Determination of bone age by paediatricians has remained one of the most typical indications that require the use of radiology, and the historical method by which radiologists determine bone age is from a bone age atlas. However, there are still some challenges. The limited case diversity related to race, geography, and age distribution; small sample sizes; and the lack of expert validation by multiple radiologists limit the generalisability of current models. Many underlying comorbidities or health histories are often overlooked in developed models. The models of the future, which provide greater accuracy and clinical usefulness, must encompass more diverse datasets, more thorough health histories, expert validation, and fast but reliable artificial intelligence (AI) models. As an educational review, this study analysed a variety of AI-based approaches that have emerged in the past several years for paediatric bone age assessment (most using hand and wrist radiographs and often coupled with radiology reports). Of these, models such as RCNN, which we evaluated with mean absolute error, showed great potential. There are great future clinical applications and advancements that can progressively transform bone age assessment and evaluation from AI. Notably, we did identify the gaps and opportunities for potentially improving the future clinical approach of bone age assessment and evaluation.