Quantitative CT Scoring for Local COPD Severity

Abstract

Chronic obstructive pulmonary disease (COPD) is complex, and its course is difficult to predict due to its diverse pathophysiology. Small airway disease (SAD), a key component of COPD and potential target for emerging therapeutics, may be reversible in mild COPD, but left unchecked, may worsen, leading to airway loss and emphysema. The dual nature of SAD complicates clinical management of COPD patients, necessitating more accurate monitoring methods. To meet this need, we developed elastic Parametric Response Mapping (ePRM), a tiered scoring system that classifies local lung volumes by the degree of PRM-derived SAD, normal, and emphysematous tissue. In individuals with or at risk for COPD, we demonstrate that chest CT ePRM can categorize local lung tissue into distinct tiers of disease severity that distinguish between tissue characterized by early reversible SAD and progressive destruction. This level of characterization is crucial to developing personalized treatment strategies for COPD.

Keywords: chronic obstructive pulmonary disease; computed tomography of the chest; elastic principal graph; emphysema; machine learning; parametric response mapping; small airways disease.

Fig. 1:. Schematic Diagram of the ePRM workflow.

Elastic PRM (ePRM) classifies local PRM composition based on an elastic principal graph that consists of five adjoining clusters that infer progression. Each cluster is classified as a tier severity score based on distinct PRM characteristics. Tiers 0, 2, 3, and Op (terminal segments) are characterized by: elevated PRM^Norm, elevated PRM^fSAD and negligible PRM^Emph, transition of PRM^fSAD to PRM^Emph, and elevated PRM^PD (opacities that result in high attenuation areas), respectively. These tier scores are connected by a single cluster, i.e. bridge, referred to as Tier 1, which is characterized by PRM^Norm transitioning to PRM^fSAD. PRM profiles along trajectories on elastic principal graph are provided in Supplemental Fig. 2.

Fig. 2:. Longitudinal assessment of ePRM in representative subject with GOLD 2 COPD.

(a) Representative coronal slices of serial PRM, with percent volume of fSAD (yellow) and emphysema (red) provided, and ePRM are presented from a participant diagnosed with GOLD 2 and 3 COPD at baseline and 5-year follow-up, respectively. The pie charts indicate the relative contribution of fSAD within each tier at the two time points. (b) Sub-volume transition between tiers over the 5-year period is depicted in a Sankey plot. The percent volume of each tier is shown at baseline and follow-up, with the flow lines indicating the direction and quantity of flow. The width of each line is proportional to the amount of transitioning sub-volumes, with thicker lines indicating larger quantities. (c) Transition of Tier 1 sub-volumes based on tier position (Tier 1-p) are presented in a Sankey plot. Here, sub-volumes are stratified into quarters based on their relative position in Tier 1. Subject 1 had a FEV₁pp and FEV₁/FVC of 51% and 0.5, respectively, at baseline and 41% and 0.41, respectively, at year 5.

Fig. 3:. Longitudinal analyses of tier position.

Waterfall plots illustrate the changes in whole-lung mean relative position for sub-volumes that remain in the same tier over a 5-year period. Participants were separated by GOLD subgroups: No COPD (PRISm and “at-risk”), GOLD 1 and 2, and GOLD 3 and 4. X-axis is the relative number of each subject. Four panels are provided for Tiers 0 through 3. Values in the x-intercept > or < 0.5 indicate a preference in movement direction along tier. Vertical lines indicate x-intercept equal to zero.

Fig. 4:. ROC analysis of Tier 1 position to predict tier score advancement.

Receiver operator characteristic plot shows the potential of mean position at baseline for a given tier to predict the assignment of a sub-volume to a progressive tier at year 5. Outcome variable was defined as sub-volumes that progress and those that were remissive over five years. As an example, sub-volumes at Tier 1 at baseline were defined as 1 if reassigned to Tiers 2 and 3 at year 5. The remaining sub-volumes, those that remained in Tier 1 or withdrew to Tiers 0 or Op, were defined as 0. Analysis was performed separately for each COPD severity subgroup.

Fig. 5:. Evolution of sub-volume tier assignment by COPD severity.

Sankey plots are shown for subgroups No COPD (PRISm and “at-risk”), GOLD 1 and 2, and GOLD 3 and 4. The mean percent volume is provided for each tier at each time point, where flow lines indicate the transition of sub-volumes between tier assignments from year 0 to year 5. The width of each line is proportional to the amount of transitioning sub-volumes, with thicker lines indicating larger quantities. To fit in the Sankey plots, real flow line values were normalized to year 0 and year 5 relative volumes for each tier. Means and standard error of the means of the tier percent volumes and actual transition values are provided in Supplemental Table 3.