The Greulich-Pyle and Gilsanz-Ratib atlas method versus automated estimation tool for bone age: a multi-observer agreement study

Abstract

Purpose To evaluate the agreement between observers using Greulich-Pyle (GP) and Gilsanz-Ratib (GR) methods, between four specialities (radiology, pediatrics, pediatric endocrinology and pediatric radiology) and between observers and automated tool in the bone age estimation. Materials and methods A total of 99 observers participated in this questionnaire-based study. BoneXpert was used for the automated tool. Experienced, senior, and junior observers were defined by their experience, and the bone age determined by experienced observers was regarded as the ground truth. Agreement between observers was evaluated using the coefficient of variance (CV) and intraclass correlation coefficient (ICC), and they were reevaluated after adding BoneXpert to the observers. Agreement of BoneXpert, the senior, and the junior observers was also evaluated using the root-mean-square-error (RMSE) values and Blant Altman method by comparing with the ground truth. Results The CV ranged from 4.98% to 22.08%. The ICC were 0.980 for GP, 0.980 for GP and BoneXpert, 0.973 for GR, and 0.976 for GR and BoneXpert, and the ICC between four specialities ranged form 0.963 to 0.990. BoneXpert tool had the lowest RMSE values (0.504 years for GP atlas). Conclusion Automated bone age estimation showed comparable results with GP and GR methods and its utilization may decrease inter-observer variability.