Scholarlink

<p>Colorectal cancer remains a leading cause of cancer-related deaths globally, with colorectal polyps serving as primary precancerous lesions. Accurate segmentation of polyps from colonoscopy images is crucial for early diagnosis. This study introduces a Multi-Domain Fusion Polyp Segmentation Network (MDF-Polyp) that integrates spectral-domain modeling, state-space modeling, and deformable attention mechanisms. The Spectral-Domain Convolution Block (SDCB) enhances multi-scale feature representation, while the Mamba layer efficiently captures long-range dependencies. Extensive experiments on five benchmark datasets demonstrate MDF-Polyp's superior segmentation accuracy (e.g., 0.930 mDice on Kvasir) and boundary refinement, offering a robust solution for clinical polyp segmentation. The source code will be available at: <a href="https://github.com/fuyou-99/MDF_Polyp">https://github.com/fuyou-99/MDF_Polyp</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

Multi-domain fusion network for enhanced polyp segmentation in colonoscopy images

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Xiaolong Wu,
  • Gaofan Zhan,
  • Changjiang Han,
  • Wentao Zhang

摘要

Colorectal cancer remains a leading cause of cancer-related deaths globally, with colorectal polyps serving as primary precancerous lesions. Accurate segmentation of polyps from colonoscopy images is crucial for early diagnosis. This study introduces a Multi-Domain Fusion Polyp Segmentation Network (MDF-Polyp) that integrates spectral-domain modeling, state-space modeling, and deformable attention mechanisms. The Spectral-Domain Convolution Block (SDCB) enhances multi-scale feature representation, while the Mamba layer efficiently captures long-range dependencies. Extensive experiments on five benchmark datasets demonstrate MDF-Polyp's superior segmentation accuracy (e.g., 0.930 mDice on Kvasir) and boundary refinement, offering a robust solution for clinical polyp segmentation. The source code will be available at: https://github.com/fuyou-99/MDF_Polyp.