AI Model Trained on Chest X-rays Predicts Mortality Risk
A deep learning model that estimates biological age from routine chest X-rays can predict which patients face higher mortality risk, according to a study of more than 421,000 Korean adults published in Radiology: Artificial Intelligence.
Researchers used a model called AgeNet to compare each patient's "radiographic age"-an AI-derived estimate-against their actual chronological age. The analysis covered chest X-rays collected between 2006 and 2020, with follow-up data extending 8.5 years on average.
Accelerated Aging Signals Higher Risk
When radiographic age exceeded chronological age by 5 or more years, researchers classified it as accelerated aging. Over the study period, 6,500 deaths occurred, including deaths from cardiovascular disease, cancer, and respiratory illness.
Men with accelerated aging faced a 26% increased mortality risk. Women with accelerated aging faced a 52% increased risk, suggesting sex-based differences in how aging patterns affect health outcomes.
Aging Speed Matters as Much as Current Age
Among nearly 180,000 patients with at least three chest X-rays, those whose radiographic age increased rapidly showed higher mortality risk regardless of their initial health status.
Each standard deviation increase in "aging velocity"-how fast radiographic age changes year to year-corresponded to a 24% rise in mortality risk for men and 35% for women. Patients whose radiographic age increased by more than 1.5 years annually faced a 51% mortality increase for men and 71% for women.
Conversely, women with slower aging velocity (less than 0.5 years of radiographic aging per year) experienced roughly 50% lower mortality risk.
Clinical Application
Both accelerated radiographic aging and rapid aging velocity independently predicted mortality risk. The findings suggest that AI analysis of existing chest X-rays could become a screening tool in preventive medicine without requiring additional imaging.
For healthcare professionals, this work demonstrates how machine learning can extract prognostic information from images already part of routine clinical care. Learn more about AI for Healthcare applications or explore AI Research Courses to understand how models like this are developed and validated.
Your membership also unlocks:
