Scholarlink

<p>Accurate polyp segmentation methods are crucial for the early diagnosis and prevention of colorectal cancer. However, this task faces three main challenges: (1) the diversity of polyp sizes and shapes; (2) blurred boundaries between polyps and surrounding tissues; and (3) interference from image noise. To address these challenges, we propose a polyp segmentation network based on uncertainty boundary enhancement, named UBENet. UBENet is specifically designed with three core modules to tackle the aforementioned challenges. First, the Progressive Fusion Module (PFM) adopts a bottom-up multi-scale feature fusion strategy to preliminarily locate polyps and generate initial polyp masks. Second, the Local–Global Feature Extraction Module (LGFEM) jointly extracts local details and global semantic information, enhancing the model’s adaptability to polyp diversity. Finally, the Uncertainty Boundary Enhancement Module (UBEM) employs an uncertainty-guided attention mechanism to focus on blurred boundary regions, thereby further improving segmentation accuracy and mitigating noise interference. Experimental results on five public datasets, including Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, ETIS, and CVC-300, demonstrate that UBENet significantly outperforms existing mainstream methods in terms of mDice, mIoU, and other metrics, with an average mDice improvement of 1.6%. These results validate the superior performance of UBENet in polyp segmentation tasks and highlight its potential for clinical auxiliary diagnosis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

UBENet: Uncertainty Boundary Enhancement Network for Polyp Segmentation in Colonoscopy Images

  • 期刊论文

  • 发表日期：

    2026年3月27日

查看全文
  • Youhui Ye,
  • Qian Wang,
  • Zihuang Wu,
  • Hongwei Li,
  • Hua Chen

摘要

Accurate polyp segmentation methods are crucial for the early diagnosis and prevention of colorectal cancer. However, this task faces three main challenges: (1) the diversity of polyp sizes and shapes; (2) blurred boundaries between polyps and surrounding tissues; and (3) interference from image noise. To address these challenges, we propose a polyp segmentation network based on uncertainty boundary enhancement, named UBENet. UBENet is specifically designed with three core modules to tackle the aforementioned challenges. First, the Progressive Fusion Module (PFM) adopts a bottom-up multi-scale feature fusion strategy to preliminarily locate polyps and generate initial polyp masks. Second, the Local–Global Feature Extraction Module (LGFEM) jointly extracts local details and global semantic information, enhancing the model’s adaptability to polyp diversity. Finally, the Uncertainty Boundary Enhancement Module (UBEM) employs an uncertainty-guided attention mechanism to focus on blurred boundary regions, thereby further improving segmentation accuracy and mitigating noise interference. Experimental results on five public datasets, including Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, ETIS, and CVC-300, demonstrate that UBENet significantly outperforms existing mainstream methods in terms of mDice, mIoU, and other metrics, with an average mDice improvement of 1.6%. These results validate the superior performance of UBENet in polyp segmentation tasks and highlight its potential for clinical auxiliary diagnosis.