Computed Tomography Imaging-based Clusters for Chronic Obstructive Pulmonary Disease Lung Function Decline

Abstract Introduction: Cluster analysis techniques can identify subgroups with distinct characteristics and have been utilized to identify chronic obstructive pulmonary disease (COPD) phenotypes. However, it is unknown if computed tomography (CT) imaging-based clusters can identify subgroups of individuals at risk for lung function decline. Methods: Participants from the Canadian Cohort Obstructive Lung Disease study with CT imaging at baseline and clinical data collected at baseline and 3-year follow-up were investigated. Rapid lung function decline was defined as forced-expiratory-volume-in-one-second (FEV1)60mL/year. A total of 130 CT features were extracted, including: 9 conventional CT features (low-attenuation-areas-below -950HU, 15th-percentile-of-the-CT-density-histogram, low-attenuation-cluster, normalized-join-count, theoretic-airway-wall-thickness-for-10mm-lumen-perimeter, wall-area-percent, lumen-area, total-airway-count, and vessel-volume-for-vessels-less-than-5mm2/total-blood-volume) and 121 PyRadiomics features (18 first-order, 14 shape, and 75 texture features from the lung parenchyma and 14 airway shape features). Spearman correlation coefficient and principal component analysis (PCA) was used to select a subset of features for the cluster analysis. First, highly correlated features were removed (|r|>0.80) and the remaining features were used in PCA to identify which features contributed to at least 1% variance. A z-normalization was applied to the selected features. To identify the optimal number of clusters a Hierarchical clustering with the ward method was evaluated. A k-means clustering algorithm was then implemented using the optimal number of clusters identified in the Hierarchical clustering. An ANOVA was used to evaluate differences for clinical and CT imaging features between the clusters; p<0.05 was considered statistically significant. Results: 750 at-risk smokers and COPD participants were included in this study (N=262 at-risk; N=294 mild-COPD; N=173 moderate-COPD, N=21 severe-COPD). A total of 5 CT features were selected for cluster analysis, WA%, LA, a lung shape, lung texture, and airway shape feature, which identified three unique clusters. Clinical and CT imaging features were significantly different between the clusters (p<0.05). Cluster 1 (n=366) reflected a chronic bronchitis group consisting mainly of females with mild-moderate COPD with significantly reduced lung function. Cluster 2 (n=343) reflected an emphysema group consisting mainly of males with mild COPD and experienced significant lung function decline and compared to cluster 1 (p<.05). Cluster 3 (n=41) reflected a chronic bronchitis-emphysema mixed group consisting of a balance between males and females and was predominantly individuals at-risk and mild COPD. Conclusion: Overall, three unique clusters with distinct imaging phenotypes were identified and may be used for risk stratification for lung function decline, which could allow for interventions to preserve quality of life and reduce disease burden.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.