Scholarlink

<p>Colonoscopy is now a common means of detecting and diagnosing polyps, and colorectal cancer can be effectively prevented by the timely removal of polyps. In clinical practice, automated polyp segmentation technology can significantly improve the efficiency and accuracy of polyp segmentation, thus saving medical resources. However, existing polyp segmentation models still have certain limitations: (1) underutilization of multi-level encoder features, leading to loss of low-level spatial details; (2) overly complex decoders, resulting in computational inefficiency; and (3) suboptimal performance in small-target segmentation. To address these issues, this paper proposes a novel Boundary-Guided Cross-semantic Attention Cascade Network (BGACNet), which comprises three key components: (1) a shared-weight module integrating a Hybrid Convolutional Enhancement Module (HCEM) and a Spatial-Channel Synergistic Attention (SCSA) module. This design synergistically enhances local and global contextual features, narrowing hierarchical semantic gaps across spatial and channel dimensions. (2) a cross-semantic cascade attention decoder that integrates multi-level features. Notably, we introduce a Cross-Scale Selective Attention Fusion Module (CSAFM) for dynamic integration and adaptive refinement of shallow features. (3) a boundary refinement module, which consists of a multi-expansion channel refinement module (MCRM) and a residual parallel axial reverse attention module (RPA-RA), capable of capturing fine-grained information and refining boundary information at different scales. To validate the effectiveness of BGACNet, extensive experiments on five widely used benchmark datasets are conducted. The experimental results show that compared with the current mainstream polyp segmentation methods, BGACNet shows superior segmentation performance and computational efficiency, and has high value for clinical applications. The code will be publicly available at <a href="https://github.com/yzw-study/BGACNet.">https://github.com/yzw-study/BGACNet.</a></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

BGACNet: boundary-guided cross-semantic attention cascade network for polyp segmentation

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Ziwei Yang,
  • Xiaoliang Zhu,
  • Dehua Ma,
  • Hanyu Li,
  • Mengkun Li

摘要

Colonoscopy is now a common means of detecting and diagnosing polyps, and colorectal cancer can be effectively prevented by the timely removal of polyps. In clinical practice, automated polyp segmentation technology can significantly improve the efficiency and accuracy of polyp segmentation, thus saving medical resources. However, existing polyp segmentation models still have certain limitations: (1) underutilization of multi-level encoder features, leading to loss of low-level spatial details; (2) overly complex decoders, resulting in computational inefficiency; and (3) suboptimal performance in small-target segmentation. To address these issues, this paper proposes a novel Boundary-Guided Cross-semantic Attention Cascade Network (BGACNet), which comprises three key components: (1) a shared-weight module integrating a Hybrid Convolutional Enhancement Module (HCEM) and a Spatial-Channel Synergistic Attention (SCSA) module. This design synergistically enhances local and global contextual features, narrowing hierarchical semantic gaps across spatial and channel dimensions. (2) a cross-semantic cascade attention decoder that integrates multi-level features. Notably, we introduce a Cross-Scale Selective Attention Fusion Module (CSAFM) for dynamic integration and adaptive refinement of shallow features. (3) a boundary refinement module, which consists of a multi-expansion channel refinement module (MCRM) and a residual parallel axial reverse attention module (RPA-RA), capable of capturing fine-grained information and refining boundary information at different scales. To validate the effectiveness of BGACNet, extensive experiments on five widely used benchmark datasets are conducted. The experimental results show that compared with the current mainstream polyp segmentation methods, BGACNet shows superior segmentation performance and computational efficiency, and has high value for clinical applications. The code will be publicly available at https://github.com/yzw-study/BGACNet.