Background: Chronic obstructive pulmonary disease (COPD) is underdiagnosed in the community. Thoracic CT scans are widely used for diagnostic and screening purposes for lung cancer. In this proof-of-concept study, we aimed to evaluate a software pipeline for the automated detection of COPD, based on deep learning and a dataset of low-dose CTs that were performed for early detection of lung cancer.
Methods: We examined the use of deep residual networks, a type of artificial residual network, for the automated detection of COPD. Three versions of the residual networks were independently trained to perform COPD diagnosis using random subsets of CT scans collected from the PanCan study, which enrolled ex-smokers and current smokers at high risk of lung cancer, and evaluated the networks using three-fold cross-validation experiments. External validation was performed using 2153 CT scans acquired from a separate cohort of individuals with COPD in the ECLIPSE study. Spirometric data were used to define COPD, with stages defined according to the GOLD criteria.
Findings: The best performing networks achieved an area under the receiver operating characteristic curve (AUC) of 0·889 (SD 0·017) in three-fold cross-validation experiments. When the same set of networks was applied to the ECLIPSE cohort without any modifications to the trained models, they achieved an AUC of 0·886 (0·017), a positive predictive value of 0·847 (0·056), and a negative predictive value of 0·755 (0·097), which is a greater performance than the best quantitative CT measure, the percentage of lung volumes of less than or equal to -950 Hounsfield units (AUC 0·742).
Interpretation: Our proposed approach could identify patients with COPD among ex-smokers and current smokers without a previous diagnosis of COPD, with clinically acceptable performance. The use of deep residual networks on chest CT scans could be an effective case-finding tool for COPD detection and diagnosis, particularly in ex-smokers and current smokers who are being screened for lung cancer.
Funding: Data Science Institute, University of British Columbia; Canadian Institutes of Health Research.
Trial registration: ClinicalTrials.gov NCT00292552.
Copyright © 2020 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND 4.0 license. Published by Elsevier Ltd.. All rights reserved.