Scholarlink

<p>Medical image segmentation is crucial for precision medicine, yet existing methods struggle to balance global context modeling with fine-grained spatial detail. This study introduces DW-MAKNet, a U-Net-style framework that integrates Dual-Window Mamba (DW-Mamba) and Convolution-Augmented Kolmogorov–Arnold Network (CA-KAN) for robust segmentation. DW-Mamba combines traditional and center-based window scanning to preserve global and local anatomical consistency, while CA-KAN enhances nonlinear representation with convolutional receptive fields. Extensive experiments conducted on five public benchmark datasets demonstrate that DW-MAKNet consistently outperforms state-of-the-art segmentation models in terms of accuracy and robustness. In particular, ablation studies show that DW-MAKNet achieves up to a 10% improvement in Dice score on the Abdomen MRI dataset compared with transformer-based counterparts. These results highlight the effectiveness of dual-window state-space encoding and convolution-augmented KAN representation for medical image segmentation. The source code is available at: <a href="https://github.com/beginner-cjh/DW-MAKNet">https://github.com/beginner-cjh/DW-MAKNet</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

Enhanced medical image segmentation via dual-window state-space encoding and convolution-augmented KAN

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Jinhao Chen,
  • Chao Ma,
  • Jinping Chen,
  • Zhong Li,
  • Zhaoxi Fang,
  • Keli Hu,
  • Guanghua He,
  • Hancan Zhu

摘要

Medical image segmentation is crucial for precision medicine, yet existing methods struggle to balance global context modeling with fine-grained spatial detail. This study introduces DW-MAKNet, a U-Net-style framework that integrates Dual-Window Mamba (DW-Mamba) and Convolution-Augmented Kolmogorov–Arnold Network (CA-KAN) for robust segmentation. DW-Mamba combines traditional and center-based window scanning to preserve global and local anatomical consistency, while CA-KAN enhances nonlinear representation with convolutional receptive fields. Extensive experiments conducted on five public benchmark datasets demonstrate that DW-MAKNet consistently outperforms state-of-the-art segmentation models in terms of accuracy and robustness. In particular, ablation studies show that DW-MAKNet achieves up to a 10% improvement in Dice score on the Abdomen MRI dataset compared with transformer-based counterparts. These results highlight the effectiveness of dual-window state-space encoding and convolution-augmented KAN representation for medical image segmentation. The source code is available at: https://github.com/beginner-cjh/DW-MAKNet.