Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/model.json

Name

16Bit Deep Learning Bone Age Model (RSNA Pediatric Bone Age Challenge winner)

Link

https://www.16bit.ai/physis

Indexing

Keywords: Pediatrics, Hand, Convolutional Neural Network, Radiography, Bone age, Robustness, Stress testing

Content: MK, PD

RadLex: RID10345, RID39030, RID39560

Author(s)

Samantha M. Santomartino

Kristin Putman

Elham Beheshtian

Vishwa S. Parekh

Paul H. Yi

Organization(s)

Drexel University College of Medicine

University of Maryland Medical Intelligent Imaging (UM2ii) Center, Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland School of Medicine

Malone Center for Engineering in Healthcare, Johns Hopkins University

Version

1.0

Contact

Paul H. Yi, email: ude.dnalyramu.mos@iyp

Funding

Authors declared no funding for this work.

Ethical review

Retrospective study using de-identified public-domain images; considered nonhuman subjects research with IRB review waived.

Date

Updated: 2024-02-26

Published: 2024-03-13

References

[1] Santomartino SM; Putman K; Beheshtian E; Parekh VS; Yi PH. "Evaluating the Robustness of a Deep Learning Bone Age Algorithm to Clinical Image Variation Using Computational Stress Testing". Radiology: Artificial Intelligence. 2024 May;6(3):e230240.. 2024-03-13. doi:10.1148/ryai.230240. PMID: 38477660. PMCID: PMC11140516.

Model

Architecture

Convolutional Neural Network (first-place model in the 2017 RSNA Pediatric Bone Age Challenge; implementation details referenced in Appendix S1).

Availability

Browser-based application available at the time of evaluation (December 2021); later transitioned to a paid platform (https://www.16bit.ai/physis). API access was provided by the developers for batch analysis.

Clinical benefit

Automated estimation of skeletal (bone) age from pediatric hand radiographs for use alongside clinical assessment.

Clinical workflow phase

clinical decision support systems

Decision threshold

Clinical categorization used 2 SDs from chronologic age (Greulich and Pyle standards) to define normal vs delayed/advanced development (DHA dataset).

Degree of automation

Automated prediction: user uploads image and indicates patient sex; model returns predicted bone age within minutes.

Indications for use

Estimation of skeletal maturity (bone age) in pediatric patients using hand radiographs in a clinical environment.

Input

Pediatric hand radiograph image; patient sex (required by the application).

Instructions

Users upload radiograph via the application interface; indicate patient sex; receive predicted bone age. (Operational details per application at time of study.)

Limitations

Model exhibited lack of robustness to simple, clinically plausible image transformations (rotations, flips, brightness/contrast adjustments, pixel inversion, downsampling). Many transformations yielded significantly higher errors and increased proportions of clinically significant errors. The RSNA dataset lacked chronologic age, precluding CSE analysis there. The evaluated model is no longer publicly available. Results pertain to this specific 16Bit model; other architectures may differ.

Output

CDEs: RDE122, RDE123, RDE1733

Description: Predicted bone age (in months). When compared to chronologic age (if available) enables classification as normal, delayed, or advanced skeletal development by 2-SD Greulich and Pyle thresholds.

Recommendation

Exercise caution in clinical deployment; ensure physician oversight and image quality control. Consider stress testing, data augmentations, out-of-distribution detection, and uncertainty quantification prior to deployment.

Reproducibility

Study used an API from the model developers for automated batch processing; robustness results reproduced across two public datasets with scripted transformations (code and transformed images provided by authors on GitHub).

Use

Intended: Detection and diagnosis

Out-of-scope: Other

User

Intended: Radiologist, Subspecialist diagnostic radiologist