Scholarlink

<p>Medical image segmentation plays a pivotal role in advancing disease diagnosis, treatment planning and prognosis prediction through artificial intelligence. While convolutional neural networks (CNNs) and vision Transformers (ViTs) have achieved success in feature extraction for medical image segmentation, they exhibit inherent limitations—such as limited receptive fields in CNNs and high computational costs in ViTs. In this paper, we propose CorrGNet, a correlation graph-based model for 3D medical image segmentation that synergistically integrates the local feature extraction capability of CNNs with the global relational modeling power of graph neural networks (GNNs). Our model constructs graph structures via the Grapher module to capture long-range dependencies and refines feature maps using a grouped gated attention mechanism. This design achieves state-of-the-art performance while reducing both model parameters and computational overhead. Experimental results on the MSD Prostate and LiTS2017 datasets demonstrate superior performance in terms of Dice coefficient, intersection over Union (IoU) and 95% Hausdorff distance, establishing CorrGNet as an efficient and effective solution for medical image segmentation. Our code is available at <a href="https://github.com/gcy22/Segmentation">https://github.com/gcy22/Segmentation</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 graph-based attention and correlation graph convolution

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Chenyu Guo,
  • Dehua Pan,
  • Yu Du,
  • Dong Liu,
  • Yibin Huang,
  • Yaoren Zhang,
  • Ming Cong

摘要

Medical image segmentation plays a pivotal role in advancing disease diagnosis, treatment planning and prognosis prediction through artificial intelligence. While convolutional neural networks (CNNs) and vision Transformers (ViTs) have achieved success in feature extraction for medical image segmentation, they exhibit inherent limitations—such as limited receptive fields in CNNs and high computational costs in ViTs. In this paper, we propose CorrGNet, a correlation graph-based model for 3D medical image segmentation that synergistically integrates the local feature extraction capability of CNNs with the global relational modeling power of graph neural networks (GNNs). Our model constructs graph structures via the Grapher module to capture long-range dependencies and refines feature maps using a grouped gated attention mechanism. This design achieves state-of-the-art performance while reducing both model parameters and computational overhead. Experimental results on the MSD Prostate and LiTS2017 datasets demonstrate superior performance in terms of Dice coefficient, intersection over Union (IoU) and 95% Hausdorff distance, establishing CorrGNet as an efficient and effective solution for medical image segmentation. Our code is available at https://github.com/gcy22/Segmentation.