NODULe: Combining constrained multi-scale LoG filters with densely dilated 3D deep convolutional neural network for pulmonary nodule detection

Abstract

Abstract Detection of pulmonary nodules on chest CT is an essential step in the early diagnosis of lung cancer, which is critical for best patient care. In this paper, we propose an automated pulmonary nodule detection algorithm, denoted by NODULe, which jointly uses a conventional method for nodule detection and a deep learning model for genuine nodule identification. Specifically, we first use multi-scale Laplacian of Gaussian (LoG) filters and prior shape and size constraints to detect nodule candidates, and then construct the densely dilated 3D deep convolutional neural network (DCNN), which combines dilated convolutional layers and dense blocks, for simultaneous identification of genuine nodules and estimation of nodule diameters. We have evaluated this algorithm on the benchmark LUng Nodule Analysis 2016 (LUNA16) dataset and achieved a detection score of 0.947, which ranks the 3rd on the LUNA16 Challenge leaderboard, and an average diameter estimation error of 1.23 mm. Our results suggest that the proposed NODULe algorithm can detect pulmonary nodules on chest CT scans effectively and estimate their diameters accurately.