Abstract Rationale Airway abnormalities are important imaging markers of diagnosis and prognosis in COPD. Mucus plugs, linked to higher mortality even in mild disease, suggest potential for early mucolytic intervention. As visual CT assessment is subjective and time-consuming, we developed an AI-based algorithm for automated mucus plug detection and evaluated its associations with lung function, mortality, and exacerbations in the COPDGene cohort. Methods Cross-sectional data from 8,971 participants in Phase 1 of the COPDGene study were analyzed across GOLD stages 0-4 and those with preserved ratio impaired spirometry (PRISm). Airway structures were segmented from inspiratory CT scans using a 3D nnU-Net model trained on 279 subjects from the Airway Tree Modelling (ATM’22) dataset, enabling detection of visible airways disconnected from the central bronchial tree, representing potential obstructions. A lightweight three-layer convolutional neural network (CNN) encoder was trained to classify regions bridging these disjoint airway endpoints as mucus plug obstructions (Figure 1C). Mucus plug burden (total plug count) was analyzed against clinical outcomes using Spearman correlation, Cox regression, and negative binomial regression analyses, with the latter two adjusted for age, sex, race, BMI, pack-years smoked, FEV1, current smoking status, emphysema, airway wall thickness, and Pi10. Results Greater mucus plug burden was associated with lower post-bronchodilator FEV1 % predicted (ρ = -0.41, p 0.001; Figure 1A) and increased air trapping defined by low-attenuation areas (LAA -856 HU) on expiratory CT (ρ = 0.33, p 0.001). Plug burden increased with GOLD stage (Figure 1B). Participants with higher plug burden also had worse quality-of-life scores (SGRQ: ρ = 0.31, p 0.001) and reduced exercise capacity (6-minute walk distance: ρ = -0.26, p 0.001). A representative 2D multiplanar slice reconstruction from a GOLD 4 participant illustrates mucus plugs detected by the algorithm (Figure 1C, red boxes). In multivariable models among GOLD 1-4 participants, mucus plug presence was independently associated with increased mortality risk (hazard ratio = 1.28, p 0.005) and higher exacerbation frequency (incidence rate ratio = 1.32, p 0.005). Conclusion We developed an AI-based algorithm for automated mucus plug detection within the visible airway tree. Detected mucus plug burden was significantly associated with impaired lung function, air trapping, and elevated mortality and exacerbation risk. These findings support the use of AI-based mucus plug quantification as a potential imaging biomarker of disease severity and clinical outcomes in COPD. This abstract is funded by: This work was supported by the National Heart, Lung, and Blood Institute (NHLBI) of the National Institutes of Health (NIH) Grants R01 HL139690 (to CJG) and R01 HL150023 (to MKH, CJG, and CRH) and by NHLBI Grants U01 HL089897 and U01 HL089856 and NIH contract 75N92023D00011, which support the COPDGene study.
Oyer et al. (Fri,) studied this question.