Bone age determination for the 21st Century: Using AI to broaden and diversify a 60-year-old gold standard and overcome reader bias - Project Summary / Abstract Bone age (BA) is a measure of maturation of a child’s skeleton, and, as such, a key clinical indicator of growth used by pediatricians and pediatric endocrinologists. As a person’s body ages, from birth through childhood, puberty, and adulthood, the size and shape of the bones of the skeleton change. Growth plates, initially wide open, fuse progressively in childhood. The BA is meant to be the “average” age at which the skeleton reaches a certain degree of maturation. In combination with other measures, it can be used to predict future adult height and detect possible growth disorders or abnormal pubertal maturation. The estimation of BA with a radiological image of the left hand and wrist describes the degree of maturation of a child’s skeleton. The most commonly available standards for the BA estimation, such as the Greulich and Pyle (G&P) Atlas (1959) and the Tanner-Whitehouse method, involve visual inspection of X-ray images of the person’s left hand and wrist, followed by its comparison with the set of reference images. This manual inspection is not only time- consuming, but subjective, and the estimation among radiologists may vary depending upon experience / expertise. Moreover, the data collected in these approaches is outdated: the current population of the United States has been much reshaped in the last 60 years, due to a larger number of children of international ancestry and a nutritional environment, particularly during the COVID pandemic, that has resulted in an obesity pandemic that affects the growth and rate of physical maturation of children. Thus, there is a need to renovate these bone age standards such that the reference is representative of the current population and develop an AI-assisted system for the accurate prediction of bone age. Moreover, the research team for the proposed project will develop an adjustment factor to incorporate the BMI Z-score for more clinical relevance of the AI- assisted outcome. The proposed technical approach is three-pronged. Specific Aim 1 is to establish a database for BA assessment in children that addresses racial and ethnic disparities and reduces spacing between available standards. Specific Aim 2 is to develop and validate an AI-assisted classification system for BA readings from the X-ray images. Finally, Specific Aim 3 is to enhance BA determination by an adjustment factor reflecting the impact of BMI Z-score on skeletal maturation. In support of all three aims, a web designer will build and deploy a web application capable of incorporating the AI algorithm for the adjusted Bone Age determination with qualitative data provided by users.
