Automatic Nodule Segmentation Method for CT Images Using Aggregation-U-Net Generative Adversarial Networks

Nodule segmentation plays a vital role in the detection and diagnosis for lung cancer. Nevertheless, manual segmentation by radiologists can be time-consuming and labor-intensive. In recent years, deep learning methods have performed well on medical image segmentation. Generative adversarial networks (GAN) are often used for image generation. In this paper, GAN is introduced to perform image segmentation, and a network called Aggregation-U-net GAN (AUGAN) is proposed, which is applied to automatically segment nodules in chest computed tomography (CT) images. The generator, which is modified by combining U-net with deep aggregation, learns features of lung nodules and makes the segmentation close to the ground truth. We used a method called tissue augmentation that 300 patches from normal areas in chest CT scan of normal lungs were selected manually and superimposed over lesions. The final results showed that our approach is superior to others for nodule segmentation and its dice coefficient and hausdorff distance were significantly improved.