Quantitative Analysis Of Emphysema From Whole-Lung, Low-Dose Computed Tomography Images

Abstract

Pulmonary emphysema is an irreversible disease of the lungs characterized by the destruction of the alveolar air sacs. Given the nature of the disease, it becomes important to be able to accurately quantify disease state in order to track progression. The advent of computed tomography has allowed for quantification of the anatomical basis of the disease, and multiple densitometric measures have been proposed for the quantification of emphysema from CT. However, two primary issues are common to density-based image scores of emphysema and remain unresolved: measure variability and poor correlation to pulmonary function test (PFT) scores. In this body of research, four primary measures have been implemented: the emphysema index, the fractal dimension, the n-th percentile of the histogram, and the mean lung density. While all have been proposed as measures of emphysema from CT, limited work has been done to analyze these measures for their validity in measuring emphysema progression, due to the variability inherent in these measures, due to inspiration variation and inconsistent scan acquisition parameters. In order to reduce this inter-scan variability, an inspiration-compensation method for reducing inter-scan measure variation based on multivariate modeling of the relationship between inspiration and metric change was developed and evaluated. Application of this system on a short time-interval longitudinal dataset was able to improved metric repeatability, with up to 45% reduction in metric variation depending on measure. This shows that inspiration compensation is possible and should be applied to future longitudinal studies of emphysema. In addition to metric variation, density-based image scores of emphysema are known to poorly correlate (r<0.5) with pulmonary function test scores, the gold standard in clinical assessment of emphysema by pulmonologists. This is particularly true with regards to relatively asymptomatic patients. To address this issue, a geometry-based, diaphragm curvature assessment of emphysema severity was implemented to take advantage of an associated symptom of emphysema: hyperinflation. The diaphragm curvature measure correlated with pulmonary function tests (r=0.24) and gas-diffusion measures (r=0.57). In addition, the geometry score did not correlate with emphysema index or fractal dimension (r<0.1), indicating that the new score provides information on disease state distinct from what is given by the densitometric measures. Multivariate modeling incorporating various image scores of COPD severity to predict pulmonary function test scores managed to further improve these findings, with a final correlation of r=0.54 between spirometric PFT scores and image-derived predicted values for non-severe stage COPD patients, who are often asymptomatic, and thus of relevant clinical interest.