Imaging-based assessment of lung function in a population cooking indoors with biomass fuel: a pilot study

Abstract

Biomass fuels (wood) are commonly used indoors in underventilated environments for cooking in the developing world, but the impact on lung physiology is poorly understood. Quantitative computed tomography (qCT) can provide sensitive metrics to compare the lungs of women cooking with wood vs. liquified petroleum gas (LPG). We prospectively assessed (qCT and spirometry) 23 primary female cooks (18 biomass, 5 LPG) with no history of cardiopulmonary disease in Thanjavur, India. CT was obtained at coached total lung capacity (TLC) and residual volume (RV). qCT assessment included texture-derived ground glass opacity [GGO: Adaptive Multiple Feature Method (AMFM)], air-trapping (expiratory voxels ≤ -856HU) and image registration-based assessment [Disease Probability Measure (DPM)] of emphysema, functional small airways disease (%AirTrap_DPM), and regional lung mechanics. In addition, within-kitchen exposure assessments included particulate matter <2.5 μm(PM_2.5), black carbon, β-(1, 3)-d-glucan (surrogate for fungi), and endotoxin. Air-trapping went undetected at RV via the threshold-based measure (voxels ≤ -856HU), possibly due to density shifts in the presence of inflammation. However, DPM, utilizing image-matching, demonstrated significant air-trapping in biomass vs. LPG cooks (P = 0.049). A subset of biomass cooks (6/18), identified using k-means clustering, had markedly altered DPM-metrics: greater air-trapping (P < 0.001), lower TLC-RV volume change (P < 0.001), a lower mean anisotropic deformation index (ADI; P < 0.001), and elevated % GGO (P < 0.02). Across all subjects, a texture measure of bronchovascular bundles was correlated to the log-transformed β-(1, 3)-d-glucan concentration (P = 0.026, R = 0.46), and black carbon (P = 0.04, R = 0.44). This pilot study identified environmental links with qCT-based lung pathologies and a cluster of biomass cooks (33%) with significant small airways disease.NEW & NOTEWORTHY Quantitative computed tomography has identified a cluster of women (33%) cooking with biomass fuels (wood) with image-based markers of functional small airways disease and associated alterations in regional lung mechanics. Texture and image registration-based metrics of lung function may allow for early detection of potential inflammatory processes that may arise in response to inhaled biomass smoke, and help identify phenotypes of chronic lung disease prevalent in nonsmoking women in the developing world.

Keywords: environmental exposure; functional small airways disease; hut lung; image registration; quantitative computed tomography.

Graphical abstract

Figure 1.

Flow chart of subject recruitment and study participation. Thirty-four subject homes were identified and recruited for this study. A subset of 25 subjects, 20 from biomass homes and 5 from LPG homes, underwent lung function assessment with spirometry and CT imaging. Two biomass subjects did not complete the study. LPG, liquified petroleum gas.

Figure 2.

t-MPR view of CT images and corresponding DPM label maps from one representative LPG cook (top row) and biomass cook (bottom row). Previously called Hyperion view (VIDA Diagnostics, IA), the t-MPR view provides an optimized visualization of the complete airway tree in the context of its surrounding tissue by warping the airway tree and its associated parenchyma onto a single plane. The t-MPR views correspond to coregistered CT data from full-inspiration (TLC) and full-expiration (RV) image data sets used in the DPM assessment. Images at both volumes are of the same scale and were cropped to center on the lung field of view. Visual inspection of the images shows an elevated lung density at TLC in the biomass subject’s lungs in contrast to the greater diaphragm shift and the lung density increase between TLC and RV in the LPG subject, indicating air-trapping in the biomass subject. The first DPM label identifies most of the LPG lung as normal (green), whereas a significant fraction of the lung exposed to biomass is classified as air-trapped (yellow). The second map depicts the regional Jacobian, representing the regional volume change in the lung, with values in red showing a local Jacobian of ≤ one, and green depicting regions with high inflation, i.e. Jacobian ≥ 2.5. The last color map shows the distribution of ADI across the lung on a color scale from zero (red) to ≥ one (green). ADI, anisotropic deformation index; DPM, disease probability measure; LPG, liquified petroleum gas; RV, residual volume; TLC, total lung capacity; t-MPR, topographic-multiplanar reconstruction.

Figure 3.

Identification of a “vulnerable” cluster of subjects based on image registration-based lung function metrics. Red circles correspond to subjects identified using k-means clustering who belong to a distinct cluster based on imaging features. Remaining biomass subjects and LPG users are shown in orange and blue circles, respectively. The circle size is scaled to air-trapping measured using the DPM approach. ADI, anisotropic deformation index; DPM, disease probability measure; LPG, liquified petroleum gas.

Figure 4.

Parenchymal texture labeling. The texture labels are overlaid onto a coronal slice from inspiratory image data sets. The three panels show results from one LPG cook (panel 1), a noncluster biomass cook (panel 2), and a biomass cook identified as belonging to the “vulnerable” cluster (panel 3). Texture labeling was performed using AMFM on the inspiratory CT data sets. Voxels in green show presence of ground glass opacities, and voxels in blue represent reticulated ground glass opacities. Labels corresponding to bronchovascular texture (which follow the branching patterns of airways and vessels) are not displayed for better visualization of the ground glass textures. Subjects in the LPG and cluster data sets are the same as from previous figures. AMFM, Adaptive Multiple Feature Method; LPG, liquified petroleum gas.

Figure 5.

Comparison of lung density at full-expiration (RV) and full-inspiration (TLC) in Hounsfield units (HU). Compared with LPG cooks, biomass cooks in the vulnerable cluster had significantly lower lung density at RV, indicating air-trapping, and higher densities at TLC, potentially indicating the presence of inflammation due to environmental exposures. The other biomass cooks had a greater variability, but were not significantly different from the LPG group. LPG, liquified petroleum gas; RV, residual volume; TLC, total lung capacity.

Figure 6.

Air-trapping estimates using densitometric threshold and the disease probability measure (DPM). Air-trapping estimates using the −856HU threshold (A) did not indicate statistically significant differences between the three groups. In contrast, the DPM-based metrics (B) show significantly higher levels of air-trapping in all subjects, with the cluster-identified subjects showing significantly greater air-trapping compared with all other subjects. HU, Hounsfield units.

Figure 7.

Comparison of lung volume change metrics using an image registration approach. A: the mean Jacobian (mean over the whole lung) represents volume change between the coregistered lung images at full-inspiration (TLC) and full-expiration (RV). B: mean ADI (anisotropic deformation index) represents the anisotropy of regional volume change. The identified cluster of biomass cooks have smaller mean Jacobian and ADI, indicating a reduced change in lung volume between TLC and RV, and a more anisotropic regional deformation. LPG, liquified petroleum gas; RV, residual volume; TLC, total lung capacity.