An Adversarial Network Embedded with Attention Mechanism for Pancreas Segmentation

Pancreas segmentation plays an important role in the diagnosis of pancreatic diseases and related complications. However, accurately segmenting pancreas from computed tomography (CT) images tends to be challenging due to the limited proportion and irregular shape of pancreas in the abdominal CT volume. To solve this issue, we propose an adversarial network embedded with attention mechanism for pancreas segmentation in this paper. The involvement of generative adversarial network contributes to retaining much spatial information for segmentation through capturing high dimensional data distributions due to its competing mechanism between the discriminator and the generator. Furthermore, the application of attention mechanism enhances the interdependency among pixels, and thus containing contextual information for segmentation. Experimental results show that our proposed model achieves competitive performance compared with most pancreas segmentation methods.