Comparative Analysis of Bone Age Estimation Using the Manual Greulich and Pyle Method Versus Artificial Intelligence Algorithms in an Indian Pediatric Population Aged 1-18 Years.

INTRODUCTION: Bone age (BA) estimation is an essential tool in pediatric growth assessment and endocrine evaluation. The study aimed to compare the performance and agreement of three artificial intelligence (AI)-based bone age assessment software platforms with the manual Greulich and Pyle (GP) atlas method in an Indian pediatric population. MATERIALS AND METHODS: Seventy-four hand radiographs from children aged 1-18 years were analyzed retrospectively. BA was assessed using the manual GP atlas and three automated AI software platforms: Xrayhead (AI1), 16-Bit Physis (AI2), and FreeBoneAge (AI3). Statistical evaluation included Pearson correlation, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Bland-Altman analysis to determine agreement and systematic bias between methods. RESULTS: All AI platforms showed significant correlation with the manual GP atlas (p < 0.001). Among the evaluated systems, 16-Bit Physis demonstrated the strongest agreement with GP assessment (ICC = 0.983, 95% CI: 0.973-0.989), the narrowest limits of agreement (3.53 years), and the lowest AI-related mean absolute error (1.24 years). FreeBoneAge showed good agreement (ICC = 0.903) but wider limits of agreement, whereas Xrayhead demonstrated only moderate agreement (ICC = 0.729) and produced unreliable outputs in children younger than six years. The manual GP method achieved the lowest overall mean absolute error relative to chronological age (0.76 years). CONCLUSION: 16-Bit Physis AI demonstrated the most reliable agreement with the GP atlas across all age groups and both sexes, with clinically acceptable bias and narrow limits of agreement. Xrayhead showed critical limitations in young children.