Optimized detection of pulmonary emphysema on CT scans of lung cancer screening participants

Abstract

Introduction: Visual CT emphysema scoring has been shown to improve the prediction of lung cancer mortality and COPD death, but optimal Hounsfield unit (HU) threshold for emphysema detection has not been determined for lung cancer screening (LCS) subjects. Objectives: To find the optimal HU threshold for the early detection of emphysema in a low dose CT (LDCT) LCS dataset. Methods: 352 participants were selected from a LCS dataset with LDCT using Definition AS, at inspiration in spiral mode at 120 kVp and 35 mAs. Images were reconstructed with B30f kernel at 2.0/1.0 mm thickness. Emphysema was quantified below a range of HU thresholds from −900 to −1,024 HU, while excluding the vessels and bronchus. One radiologist (reader A) and two trainees (reader B and C) visually scored emphysema according to the Fleischner criteria. Kappa statistics, Spearman analysis and ROC (Receiver operating characteristic) are used to assess the correlation between quantified emphysema under different HU thresholds and visual emphysema. Results: High agreement in the detection of emphysema was found (Kappa: 0.728 – 0.835). The spearman analysis demonstrated that emphysema volume under different thresholds is significantly correlated to the visual classification (P<0,05). ROC showed -1000 HU to be optimal for discrimination between normal and emphysematous lung parenchyma when correlated to visual scoring for all readers, by excluding all influence of noise on voxel quantification. Despite of noise correction, the emphysema voxel counting suffers from this threshold. Conclusions: The optimal threshold for noise cancelation in emphysema quantification was established to be -1000 HU in a LDCT LCS dataset.