Low-Dose CT Image Denoising and Pulmonary Nodule Identification

Abstract

In this paper, we describe a novel image denoising and pulmonary nodule identification method for Low-Dose CT images. Due to the decrease of the X-ray dose, LDCT images suffered from high noise and low qualities. We employ a deep convolutional neural network to not only denoise but also extract noise-free features from the noisy LDCT images. Next, these features are used to reconstruct the spatial relationship between CT slices, and we use 3D CNN to extract the spatial features. These features finally fed into a fully connected network to get the nodule identification result. The experimental results on the LUNA16 dataset show that, compared with the current deep learning algorithms, the proposed network model achieves a better sensitivity of 0.809, 0.913 and 0.945 at 1/8, 1 and 8 false positives per scan, respectively, and a higher CPM score of 0.894.