Related Experiment Video
Updated: Nov 12, 2025
Non-invasive Skeletal Muscle Quantification in Small Animals Using Micro-computed Tomography
Published on: November 8, 2024
Deep Learning to Estimate Biological Age From Chest Radiographs
Vineet K Raghu1, Jakob Weiss2, Udo Hoffmann1
1Cardiovascular Imaging Research Center, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA; Program for Artificial Intelligence in Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts, USA.
Deep learning age estimation from chest X-rays (CXR-Age) predicts mortality risk better than chronological age. This novel biological age marker improves longevity predictions and personalized care.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Longevity and Mortality Research
Background:
- Chronological age is a limited predictor of longevity and overall health.
- Biological age, reflecting health status, can enhance personalized medical care.
- A novel deep learning model estimates biological age from chest X-rays (CXR-Age) to assess mortality risk.
Purpose of the Study:
- To evaluate if deep learning-derived chest X-ray age (CXR-Age) predicts longevity beyond chronological age.
- To establish CXR-Age as a robust measure of biological age and long-term mortality risk.
- To compare the predictive power of CXR-Age against chronological age and traditional risk models.
Main Methods:
- Developed a convolutional neural network to estimate CXR-Age from chest radiograph images.
- Validated CXR-Age in two large datasets: PLCO (N=40,967) and NLST (N=5,414).
- Compared CXR-Age and chronological age in predicting all-cause and cardiovascular mortality using Cox proportional hazards models.
Main Results:
- A 5-year increase in CXR-Age showed significantly higher all-cause mortality risk (HR: 2.26) than chronological age (HR: 1.77) in the PLCO dataset.
- Similar superior predictive performance of CXR-Age was observed for cardiovascular mortality and in the NLST dataset.
- Incorporating CXR-Age into multivariable models significantly improved predictions for both mortality outcomes in both testing datasets (p < 0.001).
Conclusions:
- Chest X-ray age (CXR-Age) is a powerful deep learning-based predictor of long-term all-cause and cardiovascular mortality.
- CXR-Age offers a more accurate assessment of biological age and mortality risk compared to chronological age.
- This AI-driven approach holds potential for improving personalized risk assessment and healthcare.
