Scholarlink

With the rapid advancement of medical imaging technologies, the demand for precision medicine has been growing significantly. However, accurately delineating wound area remains challenging due to irregular boundaries, edges resembling surrounding normal skin, and complex backgrounds, which hinder the performance of existing semantic segmentation methods. To address these challenges, this study proposes an enhanced wound image segmentation method, RACL U-Net, which integrates residual connections, hybrid attention mechanism, and convolutional LSTM module to effectively capture global contextual information. The method was evaluated on a dataset provided by Beijing Jishuitan Hospital Guizhou Hospital. Experimental results demonstrate that our approach achieves an IoU of 83.08%, a Dice of 90.51%, and a HD of 6.15, outperforming existing methods in overall performance.

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

RACL U-Net: Leveraging Enhanced Global Context for Novel Wound Image Segmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xian Li,
  • Hong Luo,
  • Xiaoyan Liu,
  • Tao Zhang,
  • Mei Zhang

摘要

With the rapid advancement of medical imaging technologies, the demand for precision medicine has been growing significantly. However, accurately delineating wound area remains challenging due to irregular boundaries, edges resembling surrounding normal skin, and complex backgrounds, which hinder the performance of existing semantic segmentation methods. To address these challenges, this study proposes an enhanced wound image segmentation method, RACL U-Net, which integrates residual connections, hybrid attention mechanism, and convolutional LSTM module to effectively capture global contextual information. The method was evaluated on a dataset provided by Beijing Jishuitan Hospital Guizhou Hospital. Experimental results demonstrate that our approach achieves an IoU of 83.08%, a Dice of 90.51%, and a HD of 6.15, outperforming existing methods in overall performance.