Pulmonary nodule detection based on 3D feature pyramid network with incorporated squeeze‐and‐excitation‐attention mechanism

Abstract

Abstract Lung cancer is mainly caused by malignant lung nodules. Early detection and diagnosis of lung nodules can diagnose the disease in time and significantly improve the survival rate of the patients. With the rapid development of deep learning networks in the field of medical aid diagnosis, many deep networks have been applied to lung nodule detection. Statistical distribution shows that most of the lung nodule radii are too small to be well detected. Therefore, 3D feature pyramid network (FPN) for single‐stage pulmonary nodule detection is proposed to solve this problem by combining the 3D characteristics of computed tomography (CT) image data. In addition, the squeeze‐and‐excitation (SE)‐attention module is added to improve detection performance. The validity of the network is verified on the public pulmonary nodule dataset LUNA16. The competition performance metric (CPM) value reaches 0.8934. Compared with other pulmonary nodule detection networks, the detection performance of this network improved by 2%.