Overview

Schema Version

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

Name

CXR-Age (Deep learning–based chest radiographic age)

Link

https://dx.doi.org/10.1148/ryai.230433

Indexing

Keywords: Conventional Radiography, Thorax, Heart, Lung, Mediastinum, Outcomes Analysis, Quantification, Prognosis, Convolutional Neural Network, CXR-Age

Content: CH

RadLex: RID1243

Author(s)

Jong Hyuk Lee

Dongheon Lee

Michael T. Lu

Vineet K. Raghu

Jin Mo Goo

Yunhee Choi

Seung Ho Choi

Hyungjin Kim

Organization(s)

Seoul National University Hospital, Department of Radiology

Chungnam National University College of Medicine / Chungnam National University Hospital, Department of Biomedical Engineering

Massachusetts General Hospital Cardiovascular Imaging Research Center and Harvard Medical School

Seoul National University Medical Research Center, Institute of Radiation Medicine

Cancer Research Institute, Seoul National University

Medical Research Collaborating Center, Seoul National University Hospital

Healthcare Research Institute, Healthcare System Gangnam Center, Seoul National University Hospital

Version

1.0

Funding

National Research Foundation of Korea (NRF) grant funded by MSIT, no. RS-2023-00207978.

Ethical review

Institutional Review Board approval obtained; informed consent waived. IRB: Seoul National University Hospital H-2206-114-1335.

Date

Published: 2024-07-24

Created: 2023-10-06

References

[1] Lee JH, Lee D, Lu MT, Raghu VK, Goo JM, Choi Y, Choi SH, Kim H. "External Testing of a Deep Learning Model to Estimate Biologic Age Using Chest Radiographs". Radiology: Artificial Intelligence. 2024;6(5):e230433. 2024-07-24. doi:10.1148/ryai.230433.

[2] Raghu VK, Weiss J, Hoffmann U, Aerts HJWL, Lu MT. "Deep learning to estimate biological age from chest radiographs". JACC Cardiovasc Imaging. 2021;14(11):2226–2236. . PMID: 33744131. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11381093/

Model

Architecture

Convolutional Neural Network (as described in Raghu et al., 2021; model takes only chest radiographs as input).

Availability

Externally tested model; no code or weights provided in this article.

Clinical benefit

Prognostic risk assessment for all-cause and cause-specific (cardiovascular, lung cancer, respiratory disease) mortality using chest radiographs.

Clinical workflow phase

Clinical decision support / risk stratification during preventive health checkups or screening-related evaluations.

Degree of automation

Fully automated inference from a single PA chest radiograph; no clinical covariates required for prediction.

Indications for use

Estimation of chest radiographic age (CXR-Age) from PA chest radiographs in asymptomatic adults aged 50–80 years undergoing health checkups; intended for assessing prognostic risk for mortality outcomes.

Input

Posteroanterior chest radiograph (DICOM). No additional clinical data required for inference.

Instructions

Provide a quality PA chest radiograph; AP views should not be used. The model outputs chest radiographic age in years without needing age, sex, or smoking status as input.

Limitations

External test conducted at a single Asian center; calibration offset observed (CXR-Age higher than chronological age in this cohort); AP chest radiographs not supported; prognostic value for incident events (e.g., new CVD or lung cancer occurrence) not evaluated; clinical utility for screening selection or pulmonary function estimation not proven; certain comorbidity subgroups showed limited association with lung cancer mortality; recalibration may be required prior to clinical implementation.

Output

CDEs: RDE947

Description: Single continuous value representing deep learning–derived chest radiographic age (years).

Recommendation

Use as an adjunct to clinical factors, including chronological age, for risk stratification; not a standalone diagnostic tool.

Regulatory information

Comment: This article reports external validation only; no regulatory submission details provided.

Authorization status: Not cleared/Not a medical device in this study; research use only.

Reproducibility

Model reproducibility not assessed in this article; external performance metrics reported on an independent cohort.

Sustainability

Not reported.

Use

Intended: Risk assessment, Prognosis

Out-of-scope: Detection and diagnosis, Other

Excluded: Decision support, Procedure selection, Other

User

Intended: Radiologist, Researcher

Out-of-scope: Patient

Excluded: Layperson