Scholarlink

<p>Accurate polyp segmentation is crucial for early colorectal cancer diagnosis. This study introduces MSFFEC-Net, a novel framework integrating an edge-aware enhance module (EAEM), adaptive feature interaction module (AFIM), adaptive semantic-aware hierarchical fusion module (ASHFM), and contrastive learning to address challenges including boundary ambiguity, size variability, and camouflaged lesions. Experimental results on five public datasets demonstrate superior performance, with MSFFEC-Net achieving a 2.1% improvement in Jaccard index and a 1.7% improvement in Dice coefficient over state-of-the-art methods on CVC-ClinicDB. The proposed framework also exhibits excellent generalization ability and real-time processing speed of 50.60 FPS with only 5.71G FLOPs, making it suitable for clinical applications. The source code is publicly available at: <a href="https://github.com/Lxycherryup/MSFFEC-Net">https://github.com/Lxycherryup/MSFFEC-Net</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

MSFFEC-Net: enhanced polyp segmentation via multi-scale feature fusion with edge-aware enhancement and contrastive learning

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • Qiaohong Liu,
  • Xinyu Li,
  • Junshi Wang,
  • Luo Zhi,
  • Xuewei Li,
  • Tiansheng Huang,
  • Weiwei Shi,
  • Wenxia Bai

摘要

Accurate polyp segmentation is crucial for early colorectal cancer diagnosis. This study introduces MSFFEC-Net, a novel framework integrating an edge-aware enhance module (EAEM), adaptive feature interaction module (AFIM), adaptive semantic-aware hierarchical fusion module (ASHFM), and contrastive learning to address challenges including boundary ambiguity, size variability, and camouflaged lesions. Experimental results on five public datasets demonstrate superior performance, with MSFFEC-Net achieving a 2.1% improvement in Jaccard index and a 1.7% improvement in Dice coefficient over state-of-the-art methods on CVC-ClinicDB. The proposed framework also exhibits excellent generalization ability and real-time processing speed of 50.60 FPS with only 5.71G FLOPs, making it suitable for clinical applications. The source code is publicly available at: https://github.com/Lxycherryup/MSFFEC-Net.