Accelerated Aging and Aging Velocity from Deep Learning-based Chest Radiograph-derived Age for Predicting Cause-specific Mortality.

Purpose To assess the prognostic value of deep learning-derived radiographic age and aging velocity for predicting mortality in an Asian cohort. Materials and Methods This retrospective cohort study included Korean adults who underwent posteroanterior chest radiography between January 2006 and December 2020. Radiographic age was estimated using AgeNet, a deep learning model trained on the data of healthy asymptomatic individuals. Accelerated aging was defined as radiographic age exceeding chronological age by ≥5 years and aging velocity as the annual change in radiographic age from serial radiographs. Multivariable Cox and Fine-Gray models were used to estimate adjusted hazard ratios (HRs) for all-cause and cause-specific mortality. Mortality rate ratios were estimated using Poisson regression. Results A total of 421 894 adults were included (mean age, 47.4 years ± 8.8 [SD]; 227 427 [53.9%] male). During a median follow-up of 8.5 years, 6506 deaths occurred (953 cardiovascular, 3024 cancer, 1043 respiratory). Accelerated aging was associated with increased all-cause and cause-specific mortality, with stronger associations in female participants (P = .008 for interaction); the HRs for all-cause mortality were 1.26 (P < .001) for male participants and 1.52 (P < .001) for female participants. Among 179 667 individuals with three or more scans, aging velocity predicted mortality regardless of baseline status (adjusted cumulative mortality ratio per 1-SD increase: male participants, 1.24; female participants, 1.35; all P < .001). Compared with stable aging (1 year/year ± 0.5), decelerated velocity (<0.5 year/year) was associated with lower mortality risk (mortality rate ratios: male participants, 0.90, P = .18; female participants, 0.50, P < .001), whereas accelerated velocity (≥1.5 years/year) increased mortality risk (mortality rate ratios: male participants, 1.51; female participants, 1.71; both P < .001). Conclusion Radiographic age-based accelerated aging and aging velocity independently predicted all-cause and cause-specific mortality. Keywords: Conventional Radiography, Thorax, Epidemiology, Convolutional Neural Network Supplemental material is available for this article. © RSNA, 2026 See also the commentary by Babyn in this issue.