Scholarlink

<p>Lung cancer continues to be a major cause of cancer-related deaths, underscoring the importance of early detection of lung nodules. Nevertheless, accurately segmenting these nodules in CT images is challenging due to their morphological heterogeneity, indistinct boundaries, and variable sizes. This study proposes the Dynamic Tanh Shape Prior Module with Recurrent Residual Convolutional Neural Network (DSPM-R2U), an advanced segmentation model for lung nodule in CT images. The proposed model employs the Dynamic Tanh Shape Prior Module (DSPM) to replace traditional LayerNorm with Dynamic Tanh (DyT), adaptively adjusting feature map scaling through a learnable parameter <InlineEquation ID="IEq1"><EquationSource Format="TEX">\(\alpha \)</EquationSource></InlineEquation> to enhance edge feature expression. Additionally, Cross-Layer Prior Guidance leverages deep networks’ global receptive field to inform shallow feature extraction, effectively combining global semantics with local details through adaptive gating fusion. Experiments are conducted on the public LIDC-IDRI dataset. The results verify that the proposed model outperforms mainstream networks, achieving a DSC of 89.16%, IoU of 80.88%, and NSD of 97.60%, demonstrating its effectiveness in segmenting lung nodules.</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

DSPM-R2U: Dynamic tanh shape prior module with recurrent residual convolutional neural network for lung nodule segmentation in CT images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月24日

查看全文
  • Xinhua Yang,
  • Yan Li,
  • Meng Zhang,
  • Xiaodong Zhao,
  • Wufeng Liu

摘要

Lung cancer continues to be a major cause of cancer-related deaths, underscoring the importance of early detection of lung nodules. Nevertheless, accurately segmenting these nodules in CT images is challenging due to their morphological heterogeneity, indistinct boundaries, and variable sizes. This study proposes the Dynamic Tanh Shape Prior Module with Recurrent Residual Convolutional Neural Network (DSPM-R2U), an advanced segmentation model for lung nodule in CT images. The proposed model employs the Dynamic Tanh Shape Prior Module (DSPM) to replace traditional LayerNorm with Dynamic Tanh (DyT), adaptively adjusting feature map scaling through a learnable parameter \(\alpha \) to enhance edge feature expression. Additionally, Cross-Layer Prior Guidance leverages deep networks’ global receptive field to inform shallow feature extraction, effectively combining global semantics with local details through adaptive gating fusion. Experiments are conducted on the public LIDC-IDRI dataset. The results verify that the proposed model outperforms mainstream networks, achieving a DSC of 89.16%, IoU of 80.88%, and NSD of 97.60%, demonstrating its effectiveness in segmenting lung nodules.