Explainable Biological Age from Automated Chest Radiography-based Organ Quantifications: qCXR-bioage

Purpose To develop an explainable biologic age model using quantitative organ metrics from chest radiographs and evaluate the model's prognostic value for mortality. Materials and Methods In this retrospective cohort study of Korean adults (January 2006-December 2020), the authors developed a biologic age model, called qCXR-bioage, using deep learning segmentation software to extract quantitative chest radiograph biomarkers (lung area, emphysema probability, aortic diameter, heart area, and bone density). Automated measurements and XGBoost-derived clinical variables were integrated using the Klemera-Doubal method and multivariable ridge regression for 65 980 healthy adults, and the model was independently validated in 257 004 individuals. Predictive performance was assessed using R² and root mean squared error for chronological age and the Harrell concordance index (C-index) for mortality. The age gap (difference between radiographic and chronological age) was analyzed for all-cause and cause-specific mortality using multivariable Cox and Fine-Gray models. Results A total of 257 004 participants (mean age, 40.7 years ± 9.1 [±SD]; 141 778 male [55.2%]) were included. qCXR-bioage strongly correlated with chronological age (R² = 0.991 in male participants, 0.981 in female participants; root mean squared error < 2). After a median follow-up of 8.1 years, 1948 deaths were recorded. qCXR-bioage outperformed chronological age in predicting all-cause mortality (C-index in male participants, 0.769 vs 0.768 [P = .04]; C-index in female participants, 0.767 vs 0.761 [P = .001]). Accelerated aging (qCXR-bioage ≥ 0.5 year above chronological age) was associated with greater all-cause mortality (adjusted hazard ratio, 1.28 [95% CI: 1.11, 1.48; P = .001] in male participants and 1.28 [95% CI: 1.06, 1.56; P = .01] in female participants) after adjustment for chronological age and clinical confounders, with the strongest associations for cardiovascular mortality (hazard ratio, 1.73 [95% CI: 1.24, 2.41; P = .001] in male participants and 3.04 [95% CI: 1.37, 6.78; P = .006] in female participants). Conclusion Chest radiograph-based qCXR-bioage helped predict all-cause and cause-specific mortality. Keywords: Lung, Conventional Radiography, Thorax Supplemental material is available for this article. © RSNA, 2026.