Information Enhancement and Recursive Learning Network in a Coarse-Refine Manner for Pancreas Segmentation

Accurate pancreas segmentation is essential for the diagnosis and treatment of pancreatic cancer. In this paper, we propose an Information Enhancement And Recursive Learning Net-work (IERL-Net) for automatic pancreas segmentation fol-lowing a coarse-refine manner. Specifically, the Information Enhancement module (IEM) is deployed in the Coarse-Net and the Refined-Net to acquire more contextual information, which can efficiently prevent the degenerate of information. Attentively, in the Refined-Net, the Skip Attention Module (SAM) is added to IEM for optimizing the sensitivity of the network segmentation area. In addition, we design a Re-cursive Learning Module (RLM) to incorporate multi-stage visual cues. Comprehensive experiments demonstrate that IERL-Net outperforms the state-of-the-art methods.