DPCA-Net: Dual Path With 3D Channel Attention For Pulmonary Nodule Detection

Lung cancer affects human health severely and pulmonary nodule is an important manifestation of lung cancer. The diagnosis of lung cancer requires detection of each suspicious nodule. Early detection of lung cancer can greatly improve the survival rate of patients. It is very important to accurately find the location of pulmonary nodules in CT images and label them. Currently, CT is the most commonly used method to detect pulmonary nodules. In this paper, inspired by convolutional neural networks and the mechanisms of human attention, we propose a novel architectural block, which we term the Dual Path-Channel Attention (DPCA) block. This block can effectively prevent vanishing gradient phenomenon, enhance the transmission of context information and can weight related features according to the importance of different channels. We further propose a simple, highly efficient 3D pulmonary nodule detection model named DPCA-Net, which is constructed by DPCA block as a key component. Compared with other excellent pulmonary nodule detection models, our model achieves effective improvement on the public large-scale LUNA16 dataset.