This was a post-hoc analysis of a prospective clinical trial, which was registered online (http://www.chictr.org/en/; registration number ChiCTR-OCH-14004934). The Ethics Committee of the Institutional Review Board approved this study (No. 2013-114–1). All participants were fully informed of the nature of the study and provided written informed consent for participation.

In this study, we included only males to avoid factors that contribute to sex-related differences (Terada et al., 2023; Silveyra et al., 2021). Chest CT images of 1,097 subjects were retrospectively collected from August 2017 to December 2019 from an annual lung nodule screening cohort. The inclusion criteria were: 1) male subjects who underwent a non-contrast chest CT scan; 2) age between 18 and 80 years old. The exclusion criteria were: 1) subjects with a congenital deformity of the spine or thorax (n = 25); 2) subjects who had a history of lung illness or any respiratory symptoms (n = 10); 3) subjects with a history of lobectomy or pneumonectomy (n = 10); 4) subjects with noticeable respiratory or motion artifact (n = 37); 5) subjects with apparent lesions in the CT scan, such as the diffused emphysema, bronchiectasis, interstitial lung abnormality, lobar consolidation, nodules or mass of the lung, active tuberculosis or tuberculosis involving multiple pulmonary lobes, lung atelectasis, interstitial lung diseases or pleura effusion, confirmed by a more than 5-year experienced chest radiologist (n = 83); and 6) subjects with severe heart, liver, and kidney dysfunction (n = 4).

Finally, 928 male subjects were included in the analysis. All subjects were grouped by age, with 10 years as an age group, including ≤40 years old (n = 56), 41–50 years old (n = 154), 51–60 years old (n = 297), 61–70 years old (n = 277), 71–80 years old (n = 144). Figure 1 illustrates a flowchart outlining the process for selecting and grouping participants.

All CT scans were acquired with the same scanning and reconstruction conditions. CT scans were performed using a 64-MSCT scanner (Gemini TF PET/CT; Philips, Netherlands) from the apex to the base of the lungs at the end-inspiratory phase. An automatic current of 100–300 mAs (based on body weight) and a kilovoltage of 120 were used. Other scanning parameters were held constant: helical acquisition, gantry rotation time of 0.4 s, reconstructed section thickness of 1.25 mm, and reconstructed section interval of 1.25 mm. Images were reconstructed to encompass the entire lung field in a 512 × 512 pixel matrix using the full iterative reconstruction.

Raw data were stored in Digital Imaging and Communications in Medicine format and then transferred to a lung structure analysis workflow to segment the lung field and airways.

Automated computerized schemes were used to obtain whole lung field from CT acquisitions. First, the entire lung was segmented using a three-dimensional adaptive border matching algorithm (Figure 2A) (Pu et al., 2009). Pulmonary fissures were detected using computational geometry, and the surfaces of individual lobes were demarcated by representing the pulmonary fissures as implicit surface functions (Figure 2B) (Pu et al., 2011). The results of the computerized segmentation were verified by visual inspection and manually corrected when the computer failed.

Lung volume (LV) of the total lung (TL), right lung (RL), left lung (LL) and five lobes (right upper lobe (RUL), right middle lobe (RML), right lower lobe (RLL), left upper lobe (LUL), and left lower lobe (LLL)) were computed using the segmentation results by counting the voxels circumscribed by the boundaries and multiplying by the volume (mL).

Emphysema was quantified using the percentage of low-attenuation areas below −910 HU (LAA-910) (Figure 2C) and below −950 HU (LAA-950) (Figure 2D). As LAA-910 and LAA-950 were two widely used emphysema metrics (San José Estépar et al., 2025; Yang et al., 2025; Koo et al., 2023). Mean lung density (MLD) is a general indicator of all the components in a given volume of lung, which includes the air, acini, small airways, arteries, veins, capillaries, and epithelial lining fluid. MLD can be quantified by the average density of all the voxels. The ratios of the LAA-910 to LV and LAA-950 to LV were recorded as LAA-910% and LAA-950%, respectively.

The airway tree is automatically recognized and extracted by a geometric algorithm (Pu et al., 2012). The number of bronchi (NB) and the volume of bronchi (VB) were also calculated for TL, RL, LL, and five lobes (Figures 2E,F).

The degree of mismatch between the airway and lung, an indicator of lung function in healthy subjects, is defined as the ratio of airway volume to lung volume (ALR) on a CT scan for TL, RL, LL, and five lobes (Maetani et al., 2023).

The following statistical analyses were conducted using IBM SPSS Statistics version 25.0 (IBM Corp., Armonk, NY, United States). Baseline characteristics and quantitative parameters of the lung structures were summarized by age group, using mean ± standard deviation (SD) for normally distributed variables and median (interquartile range) for skewed variables. For variables with a normal distribution, one-way analysis of variance (ANOVA) was used to assess differences among groups; for variables without a normal distribution, the Kruskal-Wallis test was applied. The least significant difference (LSD) for variables with equal variance or Tamhane’s T2 for variables with unequal variance was then used to compare the two groups. A paired t-test or paired samples Wilcoxon Signed-Rank test was used to evaluate the quantitative parameter differences between the right and left lung. The main structural lung changes between one age stage and the previous age stage were identified using binary logistic regression. A two-sided P-value less than 0.05 was considered statistically significant for all statistical analyses.

The following statistical analyses were conducted using the MATLAB (R2023a, The MathWorks, Inc.). A polynomial weighted least squares fitting method incorporating the RLOESS smoothing technique was fitted to visualize the changing trend of quantitative parameters throughout lung aging, where the weighting scheme is adaptively determined through an iterative robust estimation process.

The LungAge model was conducted using R language (version 4.0.5). The correlation analysis was used to initially select correlated parameters. Then, the generalized additive model (GAM), a flexible nonlinear model that extends the generalized linear model by replacing linear predictors with additive functions, was applied to select the most discriminative quantitative parameters. Then, a machine learning model method, eXtreme Gradient Boosting, was used to build a predictive model. Then, mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) were calculated to evaluate the model, and a 5-fold cross-validation was applied to test its performance by R². Finally, the potential confounders, height, weight, smoking history, and BMI, were analyzed.

Totally, 928 males from an annual lung nodule screening cohort aged 25–85 were divided into five subgroups. The height and weight of71–80 years-old were significantly lower than those of the former three groups (P < 0.05). The BMI were not significantly different among the five age groups (P > 0.05).

For the TL, RL and LL, LV peaked in the group of 51–60 years-old. The LV of the TL and RL of 51–60 years-old was significantly larger than that of ≤40 and 41–50 years-old (P < 0.05) (as shown in Table 1).

For the TL, RL and LL, NB peaked at 61–70 years-old. The NB in the 51–60 and 61–70 years-old age groups was significantly higher than that in the ≤40 years old group (P < 0.05) (as shown in Table 1).

The VB of TL peaked in the group of 71–80 years-old, and VB of RL and LL peaked at 61–70 years-old. For the TL, RL and LL, the VB of 61–70 years-old was significantly higher than that of the ≤40 and 41–50 years-old (P < 0.05) (as shown in Table 1).

MLD decreased with age, while LAA-910, LAA-950, LAA-910%, LAA-950%, and ALR all showed an increasing trend with age. MLD was significantly lower in the 71–80 years-old group than that of the ≤40 years-old, 41–50 years-old, and 51–60 years-old groups, while LAA-910%, LAA950%, and ALR of 71–80 years-old were significantly higher than those of the other four age groups (P < 0.05) (as shown in Table 1).

In all five age groups, LV, NB, and VB of the RL were larger than those of the LL (P < 0.05). MLD, LAA-950, LAA-950%, and ALR of the RL were lower than those of the LL (P < 0.05) (as shown in Table 2).

The LV of bilateral upper lobes increased with age, while a decline of LV of bilateral lower lobes was observed since the age group of the sixties. The significance was observed in theRLL between the age groups of 61–70 years-old and 51–60 years-old, as well as between the 71–80 years-old and 51–60 years-old groups (P < 0.05) (as shown in Table 3).

VB and NB in all five lobes of the lungs increased gradually with age and peaked in the group of 61–70 years. The NB and VB in all five lobes of the 61–70 years-old were greater than those of the ≤40 years-old significantly (P < 0.05) (as shown in Table 3).

MLD of the bilateral upper lobes and the RML was diminished with age, and the significance was observed between the group of 71–80 years-old and the groups of ≤40 years-old, 41–50 years-old, and 51–60 years-old (P < 0.05). The MLD of the bilateral lower lobes decreased, and a significant difference was observed between the groups of 71–80 years-old and 51–60 years-old (P < 0.05) (as shown in Table 3).

LAA-910, LAA-950, LAA-910%, LAA-950%, and ALR increased gradually with age in all five lobes (all P < 0.05). The LAA-910%, LAA-950%, and ALR of the 71–80 years-old in all five lobes were higher than those of the other four groups (P < 0.05) (as shown in Table 3).

The ridge plot was used for illustration of age-related lung changes at the lobe level, see as Figure 3.

The fitted curves helped to reveal the trajectories of the quantitative parameters (see Figure 4). LV of TL, RL, LL and LUL peaked at about 60 years-old. LV of RUL exhibited a monotonic increase without decline, whereas LV of bilateral lower lobes demonstrated progressive reduction. Both VB and NB reached their peak values between 61-70 years of age.

The MLD showed a rapid decline from the age of 30–40 years-old, remained stable or slightly increased during 40–50 years-old, then began to decline again after 50 years-old. A final turning point occurred at 70 years-old, after which MLD exhibited a slight upward trend. However, the MLD of the RML consistently demonstrated a decreasing trend throughout.

LAA-910 and LAA-910% exhibited a mild decrease during ages 30–40, followed by a sustained increase until 60 years, after which a rapid increase and a downward trend were observed. LAA-950 and LAA-950% of TL decreased during ages 30–50, and then progressively increased with aging, and a similar trend of LAA-950 and LAA-950% can be seen in LL, RUL and LUL. However, the LAA-950 and LAA-950% of RLL and LLL displayed a steeper increase began at 60 years old, and finished at the 71–80 years old.

We compared the differences in the quantitative parameters between the two adjacent age groups. Using binary logistic regression, we identified the most significant changes between one age stage and the previous age stage, as shown in Table 4. The increase in ALR of LLL was the most significant characteristic between the age group ≤40 and 41–50 years old [OR = 1.315, 95% CI (1.094, 1.580), P = 0.003]. The most important lung structural changes from the age of 51–60 years-old to the age of 61–70 years-old pointed to LAA-910 of LLL and LAA-910% of LLL. The most important lung structural changes from the age of 61–70 years-old to the age of 71–80 years-old were the emphysema indexes.

Spearman correlation was firstly used to identify the related parameters during lung aging (see Supplementary Table). The selected parameters in the GAM and their importance were shown in Figure 5A. The fitted value vs. the response value were shown in Figure 5B. The MSE, RMSE, MAE, and MAPE were 89.789, 9.475, 7.478, and 13.76%. In the 5-fold cross-validation, the average R² in the training group was 0.408, the average R² in the testing group was 0.148, see as Figure 5C. The residual plot showed normal distribution, see as Figures 5D,E. Finally, the potential confounders, smoking intensity, height, weight, and BMI, were analyzed to determine their effect on the LungAge score, as shown in Figures 5F–I.