Boundary-Guided Refinement Strategy for Polyp Image Segmentation

Polyp image segmentation is critical in the early detection and diagnosis of colorectal cancer, a major health issue worldwide. Accurate segmentation of polyps in colonoscopy images is essential for identifying potential cancerous growths. In this paper, we propose a novel method for polyp image segmentation, named Polyp-BG, aimed at enhancing segmentation accuracy by integrating boundary features to guide feature refinement. The proposed method introduces two key modules: the Position Module (PM) and the Boundary-Guide Module (BGM). The PM progressively refines multilevel features to extract rough boundary information. Then, the BGM leverages these boundary features to further refine the multilevel features, improving the segmentation results. Finally, up-sampling is applied to restore feature resolution, producing the final segmentation map. We evaluate our method on several polyp image datasets, and the results demonstrate its superior performance in achieving accurate segmentation. Therefore, our proposed approach may provide valuable insights and direction for future research in polyp segmentation and could be beneficial for clinical applications in colorectal cancer detection.