Scholarlink

<p>Semantic segmentation is crucial for efficient medical image analysis, enhancing diagnostic accuracy and efficiency. However, existing networks often suffer from limited feature interaction, inadequate attention mechanism integration, and isolated deep-level semantic information. To address these issues, we propose TlED-Net, a novel semantic segmentation network based on a triple-loop encoder-decoder architecture (TlEDA) with dense skip connections. TlED-Net employs different representation strategies across its network depth, optimizes different shallow architectures, and introduces innovative modules such as dense atrous spatial pyramid pooling (DASPP) and channel space mixed attention. Experimental results on six medical datasets demonstrate that TlED-Net outperforms 17 state-of-the-art models, achieving Dice coefficients of up to 97.145% on the LUNG dataset and 93.176% on the skin lesion dataset. These findings highlight the effectiveness of TlED-Net in extracting and interpreting complex semantic information, positioning it as a valuable tool for medical image analysis.</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

TlED-Net: optimizing semantic segmentation via triple-loop encoder-decoder architecture with dense skip connections

  • 期刊论文

  • 发表日期：

    2025年6月29日

查看全文
  • Yuanhong Wei,
  • Yuefei Wang

摘要

Semantic segmentation is crucial for efficient medical image analysis, enhancing diagnostic accuracy and efficiency. However, existing networks often suffer from limited feature interaction, inadequate attention mechanism integration, and isolated deep-level semantic information. To address these issues, we propose TlED-Net, a novel semantic segmentation network based on a triple-loop encoder-decoder architecture (TlEDA) with dense skip connections. TlED-Net employs different representation strategies across its network depth, optimizes different shallow architectures, and introduces innovative modules such as dense atrous spatial pyramid pooling (DASPP) and channel space mixed attention. Experimental results on six medical datasets demonstrate that TlED-Net outperforms 17 state-of-the-art models, achieving Dice coefficients of up to 97.145% on the LUNG dataset and 93.176% on the skin lesion dataset. These findings highlight the effectiveness of TlED-Net in extracting and interpreting complex semantic information, positioning it as a valuable tool for medical image analysis.