Although a patient's chronological age is considered a risk factor when predicting lung cancer survival, the measure isn't perfect due to variance in aging rates. To see if prognostication could be improved by an AI-based estimate of biological age on routine chest x-rays, researchers from Massachusetts General Hospital and Brigham and Women's Hospital developed CXR Chest-Age.
After training CXR Chest-Age on 147,497 chest radiographs of 40,643 asymptomatic subjects in a lung cancer screening trial, the researchers then compared the prognostic power of the deep-learning algorithm with that of chronological age in 5,414 heavy smokers in the National Lung Screening Trial, as well as 604 patients with histologically confirmed lung cancer in the Boston Lung Cancer Study.
Presenter Dr. Jakob Weiss of Brigham and Women's Hospital and colleagues found that the AI-estimated biological age produced significantly higher correlation with lung cancer deaths than chronological age. What's more, adding CXR Chest-Age results to a multivariate Cox model with demographic and clinical risk factors yielded a significant improvement in estimating lung cancer-specific survival in patients from the Boston Lung Cancer Study.
"Deep learning can automatically extract prognostic information from a routine chest radiograph and improve prognostication in screening and cancer populations, which may guide decision making and personalize patient management," the authors wrote.
Exactly how well did CXR Chest-Age perform? Stop by this talk on Monday morning to find out.