COPDxNet: An End-to-End Deep Neural Network for COPD Detection

摘要: Rationale Approximately 392 million people worldwide and 24 million in the United States (US) have chronic obstructive pulmonary disease (COPD). Of these, 95% globally and 75% in US remain undiagnosed. Spirometry remains significantly underutilized. CT scans of the chest are frequently acquired in clinical practice and may facilitate opportunistic screening. Prior deep neural networks developed for the detection of COPD were limited by to the requirement for multiple preprocessing steps or need for expiratory images, which are not often acquired in clinical practice. We developed an inspiratory CT-based end-to-end convolutional neural network (CNN) to detect COPD (COPDxNet), with no preprocessing steps. We evaluated the model performance in both standard-and low-dose CT images in a large multicenter cohort. Methods COPD was defined by airflow obstruction on post-bronchodilator spirometry …