Scholarlink

<p>Automatic polyp segmentation technology is important for the early detection and treatment of colorectal cancer. Although deep learning-based methods have achieved significant progress, they still suffer from the <i>detail information neglect problem</i>, resulting in insufficient segmentation of polyps with obscure borders and overlooking small polyps, which severely affects segmentation accuracy. To overcome this problem, a novel deep neural network called Details Restoration Networks (DeRe-Net) has been proposed. In our DeRe-Net, a Details Saliency Enhancement Module (DSEM) is designed to collect and enhance a large amount of detail information. Then, the developed Feature Seamless Integration Decoder (FSID) takes responsibility for seamlessly integrating the detail features and the high-level features. Lastly, the final polyp segmentation results are obtained by the Selective Feature Aggregation Module (SFAM), which adaptively selects the related details and highest-level features to aggregate. Extensive experiments demonstrate that our DeRe-Net achieves superior segmentation performance compared to the current state-of-the-art methods on five rigorous datasets. Notably, it increases the mean Dice value of the baseline method for segmenting small polyps by 15.6%.</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

DeRe-Net: details restoration networks for polyp segmentation

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Huan Wan,
  • Qinqin Wang,
  • Jinshan Zeng,
  • Xin Wei

摘要

Automatic polyp segmentation technology is important for the early detection and treatment of colorectal cancer. Although deep learning-based methods have achieved significant progress, they still suffer from the detail information neglect problem, resulting in insufficient segmentation of polyps with obscure borders and overlooking small polyps, which severely affects segmentation accuracy. To overcome this problem, a novel deep neural network called Details Restoration Networks (DeRe-Net) has been proposed. In our DeRe-Net, a Details Saliency Enhancement Module (DSEM) is designed to collect and enhance a large amount of detail information. Then, the developed Feature Seamless Integration Decoder (FSID) takes responsibility for seamlessly integrating the detail features and the high-level features. Lastly, the final polyp segmentation results are obtained by the Selective Feature Aggregation Module (SFAM), which adaptively selects the related details and highest-level features to aggregate. Extensive experiments demonstrate that our DeRe-Net achieves superior segmentation performance compared to the current state-of-the-art methods on five rigorous datasets. Notably, it increases the mean Dice value of the baseline method for segmenting small polyps by 15.6%.