Extraction of the subpleural lung region from computed tomography images to detect interstitial lung disease.

Abstract

To quantify lesions in the subpleural lung region (SubPL) on computed tomography (CT) images and to evaluate whether they are useful for detecting interstitial lung disease (ILD). The subjects were 40 patients with idiopathic pulmonary fibrosis (IPF) diagnosed by multidisciplinary methods and 35 age-matched patients without ILDs. The lungs and SubPL were extracted from CT images using a Gaussian histogram normalized correlation system and evaluated for the mean CT attenuation value (CTmean) and the percentage of high attenuation area (%HAA), exceeding -700 Hounsfield units. The H pattern was defined as a honeycomb appearance and/or fibrosis with traction bronchiectasis, and the H-pattern volume ratios for the whole lung and the 2-mm-wide SubPL were measured. The utility of the SubPL for detecting ILD was evaluated by receiver operating characteristic (ROC) analysis. The areas under the ROC curves (AUCs) of CTmean and %HAA for the SubPL were greater than those for the whole lung. The AUCs for the whole lung and the SubPL were 0.990 and 0.994, respectively, for H-pattern volume; 0.875 and 0.994, respectively, for CTmean; and 0.965 and 0.991, respectively, for %HAA. The SubPL extraction method may be helpful for distinguishing patients with ILD from those without ILD.