Scholarlink

<p>Accurate classification of multi-organ cancers from histopathological images is quite a challenging task due to heterogeneous visual patterns and complex tissue structures. Most existing learning-based approaches are limited to capturing only local features and fail to accurately represent multi-scale relationships. This paper introduces the modelling of Virtual Path-Net (VPNet), a deep learning framework that utilizes a dual-graph architecture to understand both local and global spatial context. The framework adopts a joint approach of Graph Convolutional Networks (GCNs) and Transformer modules, with virtual nodes that integrate features through intra- and inter-graph interaction blocks. The proposed VPNet model is validated on four publicly available cancer datasets related to colorectal, prostate, gastric, and bladder cancers. The experimental results show that VPNet achieves higher classification accuracy and lower false-positive rates, with average improvements of 7.9% in accuracy, 8.4% in precision, 9.1% in recall, and 9.0% in F1-score compared to baseline models.</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

Virtual Path-Net: A Graph-Based Deep Learning Model for Multi-Organ Cancer Classification in Pathology Images

  • 期刊论文

  • 发表日期：

    2026年3月26日

查看全文
  • M. Lakshminarayana,
  • Basavaraj V. Hiremath,
  • Murigendrayya M. Hiremath,
  • N. Sridhar

摘要

Accurate classification of multi-organ cancers from histopathological images is quite a challenging task due to heterogeneous visual patterns and complex tissue structures. Most existing learning-based approaches are limited to capturing only local features and fail to accurately represent multi-scale relationships. This paper introduces the modelling of Virtual Path-Net (VPNet), a deep learning framework that utilizes a dual-graph architecture to understand both local and global spatial context. The framework adopts a joint approach of Graph Convolutional Networks (GCNs) and Transformer modules, with virtual nodes that integrate features through intra- and inter-graph interaction blocks. The proposed VPNet model is validated on four publicly available cancer datasets related to colorectal, prostate, gastric, and bladder cancers. The experimental results show that VPNet achieves higher classification accuracy and lower false-positive rates, with average improvements of 7.9% in accuracy, 8.4% in precision, 9.1% in recall, and 9.0% in F1-score compared to baseline models.