Scholarlink

In the field of medical image analysis, accurate segmentation of the lung airway tree is crucial to the diagnosis and treatment process, particularly for the study of respiratory diseases like chronic obstructive pulmonary disease (COPD), asthma, and lung cancer. Computed tomography (CT) technology is the main tool used to diagnose respiratory diseases. The manual inspection of medical images and diagnosis is inefficient due to limitations in manpower, time, and other resources. To increase the effectiveness of medical therapy, deep learning technology is used to automatically extract and segment the lung airway tree and to realize automated quantitative analysis of airway tree feature indices. However, the airway tree’s intricate topology and class imbalance present unique difficulties for automatic segmentation, which can have a big impact on segmentation performance. In order to improve the model’s overall perception, we present in this paper an attention-enhanced CNN and Transformer Interactive Fusion Coding Network (ESMA3D-Attention Network, ESCA-Net) that uses MedNeXt as its fundamental network structure and combines ESCA3D block coding with the original MedNeXt block coding. In the meantime, the jump connection layer incorporates the 3D Spatial Channel Attention (SCA) mechanism to improve segmentation accuracy, decrease back-ground or irrelevant structure interference, and enhance the properties of the airway’s structural areas. On the BAS dataset, ablation tests and comparisons with a few state-of-the-art models verify the efficacy of the model suggested in this work. The results of the experiments demonstrate that our approach performs better than other cutting-edge techniques.

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

ESCA-Net: Attention-Enhanced CNN and Transformer Fusion Coding Network for Lung Airway Tree Segmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xiangxiang Yao,
  • Jun Geng,
  • Weitong Xing,
  • Mengran Liu,
  • Mingzi Yuan,
  • Yuxin Xie

摘要

In the field of medical image analysis, accurate segmentation of the lung airway tree is crucial to the diagnosis and treatment process, particularly for the study of respiratory diseases like chronic obstructive pulmonary disease (COPD), asthma, and lung cancer. Computed tomography (CT) technology is the main tool used to diagnose respiratory diseases. The manual inspection of medical images and diagnosis is inefficient due to limitations in manpower, time, and other resources. To increase the effectiveness of medical therapy, deep learning technology is used to automatically extract and segment the lung airway tree and to realize automated quantitative analysis of airway tree feature indices. However, the airway tree’s intricate topology and class imbalance present unique difficulties for automatic segmentation, which can have a big impact on segmentation performance. In order to improve the model’s overall perception, we present in this paper an attention-enhanced CNN and Transformer Interactive Fusion Coding Network (ESMA3D-Attention Network, ESCA-Net) that uses MedNeXt as its fundamental network structure and combines ESCA3D block coding with the original MedNeXt block coding. In the meantime, the jump connection layer incorporates the 3D Spatial Channel Attention (SCA) mechanism to improve segmentation accuracy, decrease back-ground or irrelevant structure interference, and enhance the properties of the airway’s structural areas. On the BAS dataset, ablation tests and comparisons with a few state-of-the-art models verify the efficacy of the model suggested in this work. The results of the experiments demonstrate that our approach performs better than other cutting-edge techniques.