Shoshana Ginsburg MS, Abstract Co-Author: Nothing to Disclose

Joyce Denise Schroeder MD, Presenter: Research grant, Siemens AG

Alexander Kluiber, Abstract Co-Author: Nothing to Disclose

David Augustine Lynch MD, Abstract Co-Author: Research support, Siemens AG Scientific Advisor, Perceptive Informatics, Inc Consultant, Actelion Ltd Consultant, Gilead Sciences, Inc Consultant, InterMune, Inc

Emphysema and “smoker’s lung” (centrilobular nodularity, respiratory bronchiolitis) are characterized by changes in CT lung parenchymal texture patterns. Identification and quantification of smoking-related lung injury, before the development of emphysema, may be helpful in characterizing early disease and monitoring early treatment. This study demonstrates an automated method to quantify the extent of emphysema and/or “smoker’s lung” present in a CT scan based on image texture.

CT scans were obtained from 37 non-smokers with no emphysema (NNE), 36 smokers with no emphysema (SNE), and 34 smokers with centrilobular emphysema (CLE, GOLD stages 0-4) in the COPDGene™ Study with presence of emphysema determined visually. Training: 22 texture descriptors were computed for a total of nearly 3,000 35x35 pixel diameter regions-of-interest (ROIs) selected by an expert radiologist in 12 NNE scans, 12 SNE scans, and 12 CLE scans. These measures of texture were used to construct a logistic regression model to classify ROIs within the lung parenchyma as evidencing emphysema, “smoker’s lung” or normal lung. The resulting logistic regression model correctly classified 89.4% of normal ROIs, 74.3% of ROIs displaying “smoker’s lung”, and 94.5% of ROIs containing CLE. Testing: The model obtained during training was applied to thousands of overlapping ROIs on the remaining 25 NNE, 24 SNE, and 22 CLE scans.  

An average of 65.8% of ROIs from NNE scans were classified as normal, 30.9% as “smoker’s lung”, and 3.3% as CLE. For SNE scans classification was: 28.2% normal, 64.6% “smoker’s lung”, and 5.2% CLE. For CLE scans classification was: 20.0% normal, 27.5% “smoker’s lung”, and 52.0% CLE. Although there was overlap between normal and “smoker’s lung” textures, the patterns of “smoker’s lung” are substantially different from normal lung in never-smokers. Near-complete discrimination between CLE and normal patterns is achieved by our automated scheme.

Using the 22 texture descriptors evaluated here, we can quantify disease burden in patients with emphysema and potentially identify early smoking-related lung damage before emphysema occurs.

Early, objective identification and quantification of smoking-related lung injury, before the development of emphysema, may be helpful in preventing emphysema, by persuading patients to stop smoking.

Ginsburg, S, Schroeder, J, Kluiber, A, Lynch, D, Quantitative CT Assessment of Emphysema and “Smoker’s Lung” Using Texture Features: Automated Classification of Smoking-related Lung Injury.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11013286.html