World J Radiol. 2026 Jul 28;18(7):123894. doi: 10.4329/wjr.123894.
ABSTRACT
BACKGROUND: Chronic cardiopulmonary diseases, including chronic obstructive pulmonary disease, interstitial lung disease, and coronary artery disease, represent a major global health burden. Low-dose computed tomography (LDCT) combined with artificial intelligence (AI) quantitative imaging enables the identification of cardiopulmonary imaging abnormalities in asymptomatic individuals.
AIM: To evaluate early risk factors for cardiopulmonary imaging abnormalities in asymptomatic middle-aged and elderly population using single-inspiratory phase LDCT combined with AI-based whole-lung quantitative analysis.
METHODS: A retrospective collection was conducted on 1035 asymptomatic individuals aged ≥ 40 years who underwent routine single-inspiratory-phase LDCT screening at Zhuzhou 331 Hospital in 2025. An AI platform was utilized to automatically extract airway wall area percentage, low-attenuation area percentage, interstitial lung abnormality, and coronary artery calcification score. Based on risk stratification criteria, the population was divided into a high-risk group (n = 689) and a low-risk group (n = 346) and randomly stratified into a training set (n = 724) and a validation set (n = 311) at a 7:3 ratio. Independent sample t-test or χ 2 test was applied to analyze between-group differences. Binary logistic regression was used to screen independently associated factors, and a multivariate logistic regression model was constructed after excluding collinear variables using variance inflation factor (VIF < 5), followed by the establishment of a nomogram prediction model. The receiver operating characteristic curve and DeLong test were employed to evaluate model discrimination. The Hosmer-Lemeshow test and calibration curves were used to assess goodness of fit. Decision curve analysis (DCA) was applied to evaluate clinical net benefit, and internal validation was performed using the bootstrap method (1000 resamplings).
RESULTS: Univariate analysis showed that smoking history, abnormal metabolic status, body mass index (BMI), age, and gender were significantly associated with cardiopulmonary imaging-defined high-risk status (P < 0.05), while work style showed a marginal association in univariate analysis (P = 0.043) but did not retain statistical significance in the multivariate model (P = 0.851). After VIF collinearity screening (all VIF < 5) and multivariate logistic regression analysis with forced entry of six candidate variables, smoking history [odds ratio (OR) = 1.968 per level, 95% confidence interval (CI): 1.665-2.325, P < 0.001], abnormal metabolic status (OR = 3.266, 95%CI: 2.259-4.723, P < 0.001), BMI (OR = 1.150 per unit, 95%CI: 1.079-1.226, P < 0.001), and age (OR = 1.028 per year, 95%CI: 1.005-1.052, P = 0.018) were identified as independently associated factors for cardiopulmonary imaging-defined high-risk status, while male gender demonstrated an (inverse) association (OR = 0.427, 95%CI: 0.276-0.660, P < 0.001) after adjustment for smoking and other covariates. The nomogram-based risk stratification model integrating the above five independently associated factors achieved an area under the curve (AUC) of 0.833 (95%CI: 0.787-0.879) in the validation set, significantly outperforming the baseline model containing only age and gender (AUC = 0.647, 95%CI: 0.582-0.712; DeLong test: Delta AUC = 0.186, z = 5.256, P < 0.001). The Hosmer-Lemeshow goodness-of-fit test confirmed satisfactory calibration (training set: χ 2 = 10.40, P = 0.238; validation set: χ 2 = 6.50, P = 0.591). Calibration curves and DCA demonstrated good agreement and positive clinical net benefit. Bootstrap internal validation (1000 resamples) confirmed model robustness (mean AUC = 0.792, 95%CI: 0.760-0.824).
CONCLUSION: The LDCT-based nomogram model combined with AI quantitative imaging analysis may assist in identifying individuals with cardiopulmonary imaging abnormalities in asymptomatic screening populations, potentially guiding further diagnostic evaluation. Prospective studies are warranted to establish its role in improving clinical outcomes.
PMID:42529724 | PMC:PMC13419150 | DOI:10.4329/wjr.123894
