Scholarlink

<p>In the field of medical image segmentation, Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have been widely used as fundamental architectures. However, CNNs exhibit limitations in modeling long-range dependencies, while ViTs suffer from high computational complexity when processing high-resolution images. Recently, the emergence of State Space Models (SSMs) has provided a new solution for medical image segmentation. To this end, this paper proposes a novel medical image segmentation model, named Multi-scale Cascaded Channel Attention Visual Mamba UNet (MsC2A-VMUnet), to overcome the shortcomings of traditional methods in capturing global contextual information and parallel processing capabilities. This model integrates lightweight convolutional operations, SSMs, and the attention mechanism of ViTs, and innovatively designs the Channel Reconstruction Vision State Space (CR-VSS) block and Cascaded Attention Decoder (CASCADE). The CR-VSS block effectively reduces information redundancy and enhances feature representation capabilities through 2D-Selective-Scan (SS2D) and Channel Reorganization Module (CRM), while the CASCADE decoder leverages multi-scale features and a dual attention mechanism to improve segmentation performance. Comparative and ablation experiments were conducted on five public medical image datasets (ISBI2012, DSB2018, LIDC-IDRI, ISIC2017, LiTS), and the results demonstrate that MsC2A-VMUnet provides a new efficient and accurate solution for medical image segmentation. The detailed code of the model can be found at: <a href="https://github.com/wu501-CPU/MsC2A-VMUnet">https://github.com/wu501-CPU/MsC2A-VMUnet</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

A Multi-scale Cascaded Channel Attention Visual Mamba UNet model for medical image segmentation

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Wu Shili,
  • Qiao Yalu,
  • Qian Chao,
  • Zhang Xinyou

摘要

In the field of medical image segmentation, Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have been widely used as fundamental architectures. However, CNNs exhibit limitations in modeling long-range dependencies, while ViTs suffer from high computational complexity when processing high-resolution images. Recently, the emergence of State Space Models (SSMs) has provided a new solution for medical image segmentation. To this end, this paper proposes a novel medical image segmentation model, named Multi-scale Cascaded Channel Attention Visual Mamba UNet (MsC2A-VMUnet), to overcome the shortcomings of traditional methods in capturing global contextual information and parallel processing capabilities. This model integrates lightweight convolutional operations, SSMs, and the attention mechanism of ViTs, and innovatively designs the Channel Reconstruction Vision State Space (CR-VSS) block and Cascaded Attention Decoder (CASCADE). The CR-VSS block effectively reduces information redundancy and enhances feature representation capabilities through 2D-Selective-Scan (SS2D) and Channel Reorganization Module (CRM), while the CASCADE decoder leverages multi-scale features and a dual attention mechanism to improve segmentation performance. Comparative and ablation experiments were conducted on five public medical image datasets (ISBI2012, DSB2018, LIDC-IDRI, ISIC2017, LiTS), and the results demonstrate that MsC2A-VMUnet provides a new efficient and accurate solution for medical image segmentation. The detailed code of the model can be found at: https://github.com/wu501-CPU/MsC2A-VMUnet.