Scholarlink

<p>Accurate segmentation and classification of nuclei in hematoxylin and eosin (H&amp;E)-stained images pose significant challenges due to overlapping nuclei, blurred boundaries, and variations in staining and expression. Traditional methods rely heavily on manual feature extraction, which is time-consuming and labor-intensive. Recent deep learning-based approaches, while effective, struggle to capture robust features for precise classification and instance segmentation. In this study, we propose a novel network, HA2PNet, which incorporates a Hybrid Attention Multi-scale Feature Aggregation (HAMFA) module. The HAMFA module pioneers a hybrid attention mechanism that effectively captures cross-dimensional interactions between channel and spatial features, enabling dynamic and adaptive fusion of multi-scale contextual information. It integrates global and local features from various layers to enhance feature representation, thereby addressing morphological and staining variability. Our model achieves state-of-the-art performance on three publicly available datasets, demonstrating improved accuracy and efficiency. The lightweight design of HA2PNet, combined with the HAMFA module, enables faster training and inference speeds, making it suitable for practical applications in pathology. We also open-source our code and datasets to facilitate further research and reproducibility (<a href="https://github.com/Hedwig-XPZhang/HA2PNet-HE">https://github.com/Hedwig-XPZhang/HA2PNet-HE</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

Hybrid attention multi-scale feature aggregation for efficient nuclei segmentation and classification in H&E-stained images

  • 期刊论文

  • 发表日期：

    2025年10月16日

查看全文
  • Xingpeng Zhang,
  • Peng Guo,
  • Qiuli Wang,
  • Kaixin Wang,
  • Sijing Wu,
  • Jing Xu,
  • Yang Yu

摘要

Accurate segmentation and classification of nuclei in hematoxylin and eosin (H&E)-stained images pose significant challenges due to overlapping nuclei, blurred boundaries, and variations in staining and expression. Traditional methods rely heavily on manual feature extraction, which is time-consuming and labor-intensive. Recent deep learning-based approaches, while effective, struggle to capture robust features for precise classification and instance segmentation. In this study, we propose a novel network, HA2PNet, which incorporates a Hybrid Attention Multi-scale Feature Aggregation (HAMFA) module. The HAMFA module pioneers a hybrid attention mechanism that effectively captures cross-dimensional interactions between channel and spatial features, enabling dynamic and adaptive fusion of multi-scale contextual information. It integrates global and local features from various layers to enhance feature representation, thereby addressing morphological and staining variability. Our model achieves state-of-the-art performance on three publicly available datasets, demonstrating improved accuracy and efficiency. The lightweight design of HA2PNet, combined with the HAMFA module, enables faster training and inference speeds, making it suitable for practical applications in pathology. We also open-source our code and datasets to facilitate further research and reproducibility (https://github.com/Hedwig-XPZhang/HA2PNet-HE).