Scholarlink

<p>Pathological image detection of biopsy tissues is crucial for targeted cancer treatment and prognosis evaluation. Due to computational resource limitations, current state-of-the-art methods rely on patch-based processing to detect WSIs. Nevertheless, these approaches are resource-intensive, requiring substantial computational resources and long detection times. In this study, we argue that exhaustive patch-wise analysis of WSIs introduces substantial redundancy. Clinically, pathologists often use low-power objectives to screen WSIs for suspicious regions, switching to high-power objectives to examine these regions in detail. Inspired by this practice, we propose a multi-scale classification framework for pathological images. This framework employs the Patch Selection Module (PSM) to selectively use high-resolution analysis for WSI detection, reducing redundant holistic detection of pathological images while maintaining accuracy. Additionally, considering the multi-scale nature of WSIs, we designed the Directed Node Update (DNU) module to enable information transfer and joint analysis of the same WSI regions at different magnifications, further enhancing the reliability of detection results. To improve the interpretability of the framework, we incorporated visualization operations at multiple stages of the model, facilitating comprehensive analysis of detection results by pathologists. We conducted extensive experiments on the thyroid papillary carcinoma lymph node metastasis H&amp;E-stained dataset and the publicly available breast cancer lymph node metastasis dataset, Camelyon16. Experimental results demonstrate that our framework outperforms state-of-the-art techniques, achieving sparse detection of WSIs while ensuring patch-level analysis accuracy, significantly improving detection speed.</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 classification framework for pathological image diagnosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Peihe Jiang,
  • Yukun Xu,
  • Ming Zhang,
  • Chunni Wang,
  • Ning Lu

摘要

Pathological image detection of biopsy tissues is crucial for targeted cancer treatment and prognosis evaluation. Due to computational resource limitations, current state-of-the-art methods rely on patch-based processing to detect WSIs. Nevertheless, these approaches are resource-intensive, requiring substantial computational resources and long detection times. In this study, we argue that exhaustive patch-wise analysis of WSIs introduces substantial redundancy. Clinically, pathologists often use low-power objectives to screen WSIs for suspicious regions, switching to high-power objectives to examine these regions in detail. Inspired by this practice, we propose a multi-scale classification framework for pathological images. This framework employs the Patch Selection Module (PSM) to selectively use high-resolution analysis for WSI detection, reducing redundant holistic detection of pathological images while maintaining accuracy. Additionally, considering the multi-scale nature of WSIs, we designed the Directed Node Update (DNU) module to enable information transfer and joint analysis of the same WSI regions at different magnifications, further enhancing the reliability of detection results. To improve the interpretability of the framework, we incorporated visualization operations at multiple stages of the model, facilitating comprehensive analysis of detection results by pathologists. We conducted extensive experiments on the thyroid papillary carcinoma lymph node metastasis H&E-stained dataset and the publicly available breast cancer lymph node metastasis dataset, Camelyon16. Experimental results demonstrate that our framework outperforms state-of-the-art techniques, achieving sparse detection of WSIs while ensuring patch-level analysis accuracy, significantly improving detection speed.