Lung Nodule Detection Using Combined Traditional and Deep Models and Chest CT

摘要

Detection of lung nodules in chest CT scans is of great value to the early diagnosis of lung cancer. In this paper, we jointly use traditional object detection methods and deep learning, and thus propose a lung nodule detection algorithm for chest CT scans. We first detect all candidate nodules using multi-scale Laplace of Gaussian (LoG) filters and shape priors, and finally construct a multi-scale 3D DCNN to differentiate nodules from non-nodule volumes and estimate nodules' diameters simultaneously. This algorithm has been evaluated on the benchmark LUng Nodule Analysis 2016 (LUNA16) dataset and achieved an average diameter estimation error of 0.98 mm and a detection score of 0.913. Our results suggest that the proposed algorithm can effectively detect lung nodules on chest CT scans and accurately estimate their diameters.