Pulmonary Nodule Detection Based on Convolutional Block Attention Module

Since the diagnosis of lung cancer requires a careful investigation of each suspicious nodule, traditional detection approaches need to integrate the information of all nodules. In deep learning, the convolutional block attention module can be integrated into convolutional neural network architectures for adaptive feature refinement. Therefore, we propose a state-of-the-art pulmonary detection network, which embeds this attention module. Furthermore, the bounding box regression is redefined to enhance convergence ability. The novel detection network is validated on a public dataset from LUNA16, which achieves 0.854 sensitivity rate.