Deep Learning Can Better Assess Pulmonary Nodule Malignancy Potential

Background: Low-dose chest CT (LDCT) screening has lowered the lung cancer mortality rate but can also lead to false-positive interpretations causing overdiagnosis of potential malignancy. Considering the increasing volume of these studies, using artificial intelligence methods to accurately and efficiently interpret these examinations may be helpful. Objective: To produce a deep learning method to determine the malignancy risk of pulmonary nodules and compare this method to current malignancy risk scores. Design: Retrospective analysis. Methods: Data for developing the deep learning method called the DeepPNP (Deep Pulmonary Nodule Profiler) were obtained from 3 sources. First was the National Lung Screening Trial (NLST), which included 26,722 subjects evaluated between 2002 and 2007. This trial included a baseline LDCT and 2 annual screening LCDTs. Second was from a medical center in Europe, which included chest CTs acquired between 2010 and 2020, on patients with biopsy-proven lung more...