The authors developed MOSAIC‑Age, a computational framework that quantifies multi‑organ biological aging from routine chest CT images. MOSAIC‑Age generates eight organ‑specific aging clocks using features visible on thoracic CT to estimate each organ’s apparent biological age relative to chronological age. The approach aims to capture heterogeneous, organ‑level patterns of aging that are not fully reflected by existing global biological age measures.
Model development and internal validation used 9,971 CT scans pooled from the CT‑RATE and MIDRC datasets. After locking the models, the authors applied the pretrained organ clocks to two large independent prospective cohorts for external evaluation: the National Lung Screening Trial (NLST) and the Genetic Epidemiology of COPD (COPDGene) study, which together included 35,293 participants. The study therefore combined model training, internal validation, and independent prospective application across publicly accessible and controlled‑access imaging resources.
For each thoracic organ, MOSAIC‑Age produces a predicted biological age based on CT appearance. The CT‑derived biological age gap (BAG) is defined as the difference between the CT‑predicted organ age and the individual’s chronological age. A positive BAG indicates that an organ appears older on CT than expected for the patient’s chronological age; a negative BAG indicates an organ appearing younger. The authors examined BAGs both individually for each organ and jointly across all eight organ clocks.
Higher organ‑specific BAGs were associated with adverse health and social characteristics reported in the cohorts. Specifically, larger BAGs correlated with unfavorable lifestyle measures and socioeconomic indicators and with a higher burden of prevalent comorbidities. These associations support the construct validity of CT‑derived BAGs as markers reflecting health status beyond chronological age.
The study evaluated how BAGs related to incident chronic disease outcomes over follow‑up in the applied cohorts. Multiple disease outcomes showed associations with aging across several thoracic organs. When multivariable models included all eight organ‑specific BAGs simultaneously, many of the remaining associations became more organ specific, indicating that some disease risks map preferentially to aging in particular organs visible on chest CT.
Higher BAGs were linked to increased risk of both all‑cause and cause‑specific mortality in the evaluated cohorts. The findings indicate that organs that appear older than chronological age on routine chest CT carry prognostic information relevant to long‑term survival.
Beyond single‑organ BAGs, the authors assessed summary measures of multi‑organ aging. A greater number of markedly older‑appearing organs was associated with higher mortality. Similarly, a faster pace of aging across organs correlated with increased mortality risk. These aggregated measures suggest an additive or combinatorial signal when multiple thoracic organs show accelerated aging on CT.
The study demonstrates that routine chest CT can capture both shared and organ‑specific patterns of biological aging, proposing CT‑derived organ clocks as candidate quantitative imaging biomarkers for multi‑organ health assessment and long‑term disease risk stratification. Because chest CT is commonly performed in clinical care and screening, MOSAIC‑Age offers a potential route for opportunistic, imaging‑based phenotyping of organ health without additional procedures.
The authors note that some disease associations were nonspecific across organs, whereas others remained organ specific in multivariable analyses, highlighting opportunities to refine organ‑targeted risk prediction. They also describe availability of pretrained model weights and controlled access imaging resources to support reproducibility and further research.
Data availability for the study is described in the source: CT‑RATE imaging data are accessible via a Hugging Face repository under academic access conditions; MIDRC data are available through the MIDRC Data Commons to registered users under its data use agreement; NLST imaging and clinical data are available via The Cancer Imaging Archive and the NCI Cancer Data Access System; COPDGene imaging and phenotype data are available via controlled access on dbGaP. Paired CT/PET‑CT data are not publicly available owing to patient privacy but can be requested from the corresponding author. The pretrained organ‑clock model weights are available from the corresponding author upon reasonable request.
Institutional Review Board approval was obtained from The University of Texas MD Anderson Cancer Center. The manuscript reports multiple author disclosures and funding sources, including NIH grants and other declared industry relationships; remaining authors reported no competing interests.