Scholarlink

<p>As one of the leading causes of cancer-related deaths worldwide, timely and accurate diagnosis of colorectal cancer is paramount for effective patient management and improved survival rates. However, traditional histopathological analysis of H&amp;E-stained tissue slides remains a time-consuming and labor-intensive bottleneck. To address this, we introduce a novel automated solution for colorectal cancer diagnosis that integrates graph convolutional networks (GCNs), which effectively model complex intra-tile spatial relationships, with a quad-tree-based mask optimization technique for efficient region pruning from whole slide images (WSIs). Our method segments relevant tissue regions and then classifies these regions into nine distinct categories, enabling a more comprehensive diagnosis of colorectal cancer. It enhances classification performance by leveraging crucial intra-tile graph-based representations to capture complex tissue morphology. Furthermore, it significantly reduces computational burden by effectively eliminating irrelevant regions from WSIs. Experimental results validate the effectiveness of our proposed model, demonstrating a strong accuracy of 97.53%. Notably, our approach achieves this with up to 99.89% fewer parameters and faster processing times compared to traditional deep convolutional neural networks (CNNs) like ResNet and VGG. With an average inference time of just 9.5&#xa0;ms per tile and a lightweight model architecture, our method is exceptionally well-suited for deployment in real-time clinical settings, providing a scalable and real-time diagnostic tool that addresses both accuracy and efficiency challenges in digital pathology.</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

GCN-QuadNet: A Lightweight and Efficient Graph-Based Framework for Colorectal Cancer Segmentation in Whole Slide Histopathology Images

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Ramzi Agaba,
  • Rohallah Benaboud,
  • Mehdi Malah,
  • Fayçal Abbas

摘要

As one of the leading causes of cancer-related deaths worldwide, timely and accurate diagnosis of colorectal cancer is paramount for effective patient management and improved survival rates. However, traditional histopathological analysis of H&E-stained tissue slides remains a time-consuming and labor-intensive bottleneck. To address this, we introduce a novel automated solution for colorectal cancer diagnosis that integrates graph convolutional networks (GCNs), which effectively model complex intra-tile spatial relationships, with a quad-tree-based mask optimization technique for efficient region pruning from whole slide images (WSIs). Our method segments relevant tissue regions and then classifies these regions into nine distinct categories, enabling a more comprehensive diagnosis of colorectal cancer. It enhances classification performance by leveraging crucial intra-tile graph-based representations to capture complex tissue morphology. Furthermore, it significantly reduces computational burden by effectively eliminating irrelevant regions from WSIs. Experimental results validate the effectiveness of our proposed model, demonstrating a strong accuracy of 97.53%. Notably, our approach achieves this with up to 99.89% fewer parameters and faster processing times compared to traditional deep convolutional neural networks (CNNs) like ResNet and VGG. With an average inference time of just 9.5 ms per tile and a lightweight model architecture, our method is exceptionally well-suited for deployment in real-time clinical settings, providing a scalable and real-time diagnostic tool that addresses both accuracy and efficiency challenges in digital pathology.