EUVIMEDEUVIMED entdecken
Crossref

Validation of an AI model for bone age assessment in children and adolescents for the development of a multimodal diagnostic platform

A. Y. Surgutskaja, S. A. Smetanina, N. V. Ermakov, S. E. Kondratova, T. Y. Malysheva, A. N. Khodyrev, A. S. Belozyorov, K. S. Belozyorov, M. V. Milyaev, V. D. Tyurin, K. O. Popova, Y. L. Bezgodova und 1 weitere

Perm Medical Journal · 10.07.2026 · 43 (3) · 126-138

DOI: 10.17816/pmj433126-138 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Perm Medical Journal, online erschienen am 10.07.2026. Verfasst von A. Y. Surgutskaja und 12 weiteren Autor:innen. Die Arbeit nennt 12 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Objective

To conduct a comprehensive validation of the developed artificial intelligence model for bone age assessment using hand and wrist radiographs in children and adolescents. Materials and methods. Training a convolutional neural network for bone age regression (in months) using a dataset of hand and wrist radiographs from children and adolescents.

Results

The mean absolute error (MAE) of the model trained exclusively on the RSNA Pediatric Bone Age Challenge dataset (n = 13841) was 6.32 months. After fine-tuning the model on a sample from the PACS system of the Tyumen Regional Clinical Hospital No. 1 (n = 405), the metrics maintained comparable accuracy (MAE = 6.37 months, p = 0.795). However, a key improvement was achieved following the reannotation of the RSNA dataset: training the model on 272 reannotated images resulted in an MAE of 3.13 months. A comparative analysis of the developed model's metrics with domestic and international analogues demonstrated its alignment with the current state of research in the field of automated bone age assessment.

Conclusions

The conducted study demonstrates the high efficacy of the "Bone Age Analyzer" model (MAE = 3.13 months), which surpasses known analogues. The key factors influencing accuracy are the quality of data annotations, the representativeness of the sample, and accounting for sex-specific features of bone maturation. To further improve accuracy, the following are required: expansion of the database, implementation of consensus-based image review, and standardization of radiography protocols. The model is ready for clinical piloting and can significantly enhance the quality of growth disorder diagnosis in children.

Zitieren

APA

Surgutskaja, A. Y., Smetanina, S. A., Ermakov, N. V., Kondratova, S. E., Malysheva, T. Y., Khodyrev, A. N., Belozyorov, A. S., Belozyorov, K. S., Milyaev, M. V., Tyurin, V. D., Popova, K. O., Bezgodova, Y. L., & Sosnovtseva, E. P. (2026). Validation of an AI model for bone age assessment in children and adolescents for the development of a multimodal diagnostic platform. Perm Medical Journal, 43(3), 126-138. https://doi.org/10.17816/pmj433126-138

Vancouver

Surgutskaja AY, Smetanina SA, Ermakov NV, Kondratova SE, Malysheva TY, Khodyrev AN, et al. Validation of an AI model for bone age assessment in children and adolescents for the development of a multimodal diagnostic platform. Perm Medical Journal. 2026;43(3):126-138. doi:10.17816/pmj433126-138

BibTeX
@article{Surgutskaja2026,
  author = {Surgutskaja, A. Y. and Smetanina, S. A. and Ermakov, N. V. and Kondratova, S. E. and Malysheva, T. Y. and Khodyrev, A. N. and Belozyorov, A. S. and Belozyorov, K. S. and Milyaev, M. V. and Tyurin, V. D. and Popova, K. O. and Bezgodova, Y. L. and Sosnovtseva, E. P.},
  title = {Validation of an AI model for bone age assessment in children and adolescents for the development of a multimodal diagnostic platform},
  journal = {Perm Medical Journal},
  year = {2026},
  volume = {43},
  number = {3},
  pages = {126--138},
  doi = {10.17816/pmj433126-138}
}
Kostenlos starten →