Scholarlink

<p>Accurate boundary delineation in medical image segmentation is crucial for precise diagnoses and treatment planning. Despite significant advancements in deep learning-based technologies, the limited availability of high-quality labeled datasets poses challenges for constructing deep models with precise segmentation and good generalizability. To address these issues, we propose an efficient hierarchical multiscale convolutional attention network (EHMCANet) for medical image segmentation. EHMCANet combines hierarchical, multiscale convolutional attention and deep supervision techniques to improve segmentation accuracy. Our multiscale residual attention module effectively extracts multiscale contextual details from high-resolution features, suppressing irrelevant noise. The effective spatial–channel attention module in the network’s deep layers extracts global contexts, alleviating the semantic gap between encoding and decoding, and reducing the number of parameters. Experimental results on eleven challenging medical image segmentation datasets demonstrate that EHMCANet outperforms existing CNNs and Transformer-based models, achieving a good balance between performance and resource efficiency. Code: <a href="https://github.com/wzh0328/EHMCANet/tree/main/EHMCANet">https://github.com/wzh0328/EHMCANet/tree/main/EHMCANet</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

Efficient hierarchical multiscale convolutional attention for accurate medical image segmentation

  • 期刊论文

  • 发表日期：

    2025年7月12日

查看全文
  • Bing Wang,
  • Zhihong Wei,
  • Mengyi Ju,
  • Zutong Zhao,
  • Shiyin Zhang

摘要

Accurate boundary delineation in medical image segmentation is crucial for precise diagnoses and treatment planning. Despite significant advancements in deep learning-based technologies, the limited availability of high-quality labeled datasets poses challenges for constructing deep models with precise segmentation and good generalizability. To address these issues, we propose an efficient hierarchical multiscale convolutional attention network (EHMCANet) for medical image segmentation. EHMCANet combines hierarchical, multiscale convolutional attention and deep supervision techniques to improve segmentation accuracy. Our multiscale residual attention module effectively extracts multiscale contextual details from high-resolution features, suppressing irrelevant noise. The effective spatial–channel attention module in the network’s deep layers extracts global contexts, alleviating the semantic gap between encoding and decoding, and reducing the number of parameters. Experimental results on eleven challenging medical image segmentation datasets demonstrate that EHMCANet outperforms existing CNNs and Transformer-based models, achieving a good balance between performance and resource efficiency. Code: https://github.com/wzh0328/EHMCANet/tree/main/EHMCANet.