Research of Lung Nodule Segmentation Algorithm Based on 3D U-Net Network

Lung cancer is a common cancer that usually manifests as lung nodules in its early stages. If lung nodules can be diagnosed at an early stage and treated in time, the probability of their conversion to cancer can be effectively reduced. Based on the 3D U-Net network, this paper analyzes the current research status of lung nodule segmentation based on convolutional neural networks, and proposes a method to improve the 3D U-Net network by using Dense blocks. In addition, the structure of Deep Supervision is used to further improve the performance of the 3D U-Net network. Finally, we use the LUNA16 public dataset to compare with traditional convolutional networks on various segmentation metrics. The experimental results show that, compared with the traditional method, the improved 3D U-Net network effectively improves the efficiency of lung nodule segmentation.