Computer-aided identification of interstitial lung disease based on computed tomography

OBJECTIVE:Identification of interstitial lung disease (ILD) may be difficult in certain cases using pulmonary function tests (PFTs) or subjective radiological analysis. We evaluated the efficacy of quantitative computed tomography (CT) with 3-dimensional (3D) reconstruction for distinguishing ILD p atients from healthy controls. MATERIALS AND METHODS:We retrospectively collected chest CT images of 102 ILD patients and 102 healthy matched controls, and measured the following parameters: lung parenchymal volume, emphysema indices low attenuation area LAA910 volume, LAA950 volume, LAA910%, and LAA950%, and mean lung density (MLD) for whole lung, left lung, right lung, and each lobe, respectively. The Mann-Whitney U test was used to compare quantitative CT parameters between groups. Receiver operating characteristic (ROC) curves, Bayesian stepwise discriminant analysis, and deep neural network analysis were used to test the discriminative performance of quantitative CT parameters. Binary logistic regression was performed to identify ILD markers. RESULTS:Total lung volume was lower in ILD patients than controls, while emphysema and MLD values were higher (P < 0.001) except LAA910 volume in right middle lobe. LAA910 volume, LAA950 volume, LAA910%, LAA950%, and MLD accurately distinguished ILD patients from healthy controls (AUC >0.5, P < 0.05), and high MLD was a significant marker for ILD (OR = 1.047, P < 0.05). CONCLUSIONS:This quantitative CT analysis can effectively identify ILD patients, providing an alternative to subjective image analysis and PFTs.