[Artificial intelligence algorithm for bone age estimation].

BACKGROUND

Bone age (BA) estimation with automated methods can eliminate interindividual variation.

OBJECTIVE

To evaluate BA by creating an artificial intelligence (AI) algorithm in a pediatric population from the Bajío region of Mexico.

MATERIAL AND

METHODS

Observational, cross-sectional, retrospective, analytical study. Left-hand radiographs of children under 18 years of age, obtained from the Radiology Department database were included to create the AI algorithm for estimating BA. The BA result obtained by AI was compared with that obtained by 2 expert observers using the Greulich and Pyle method.

RESULTS

271 radiographs were analyzed to assess BA and this was similar between observers 1, 2, and AI when considering all images (p = 0.68). The time taken to estimate BA was longer with AI (p < 0.001). AI measurement showed no differences between chronological age (CA) and BA when considering both the total number of images (p = 0.12) and when they were distributed by age group: < 6 years, 6 to < 10 years, and ≥ 10 years (p = 0.60, p = 0.06, p = 0.67, respectively). The highest EO concordance correlation coefficients (CCCs) were recorded when all images were evaluated (observer 1 and 2, observer 1 and AI, and observer 2 and AI [p < 0.001, in all 3 scenarios]).

CONCLUSIONS

The AI algorithm allows for objective estimation of BA in children and adolescents as a first training approach; its refinement will optimize its use and utility in clinical practice.