Scholarlink

<p>Congenital heart disease is one of the most common diseases in children worldwide. Single ventricle congenital heart disease poses greater challenges to early diagnosis and accurate assessment due to its complex anatomical structure and unique pathological characteristics. To this end, a cascaded channel attention UNet (CAUNet) network with a channel attention mechanism for automatic segmentation of the left and right ventricles in cardiac CT images is proposed in this study, aiming at the complex anatomical structure and fuzzy boundaries of pediatric heart, to support the diagnosis and prognosis analysis of single ventricle heart disease. Compared with the traditional attention mechanism, the introduction of the channel attention mechanism can more fully capture the relationship between local and global features in the image, effectively improving the accuracy of cardiac CT image segmentation. In terms of model architecture, the designed cascaded CAUNet introduces the channel attention module in the two-stage refinement network by constructing an adaptive cascade segmentation framework; that is, the ventricular region is roughly segmented in the first stage, and the initial segmentation result is optimized in detail in the second stage. Through this two-stage segmentation strategy, the segmentation performance of the network is not only improved, but also shows strong robustness when dealing with single ventricle cases with irregular morphology. In order to verify the effectiveness of the method, an experiments on 4 real clinical data sets is conducted in this study. The experimental results show that under the premise of a small increase in computational complexity, the Dice similarity coefficient of the network proposed in this paper in the segmentation of 7 substructures of the heart, the left and right ventricles, and the pulmonary artery segmentation tasks is improved compared with the mainstream multi-organ segmentation network, proving the potential of channel attention in cardiac image segmentation applications. The obtained precise segmentation results provide accurate quantitative indicators for clinicians to assist in the diagnosis of single ventricle structural abnormalities, which has important application value.</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

Cascaded CAUNet for Single Ventricle Disease Diagnose

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Siru Liu,
  • Xu Liu,
  • Jinlong Liu,
  • Xu Zhao

摘要

Congenital heart disease is one of the most common diseases in children worldwide. Single ventricle congenital heart disease poses greater challenges to early diagnosis and accurate assessment due to its complex anatomical structure and unique pathological characteristics. To this end, a cascaded channel attention UNet (CAUNet) network with a channel attention mechanism for automatic segmentation of the left and right ventricles in cardiac CT images is proposed in this study, aiming at the complex anatomical structure and fuzzy boundaries of pediatric heart, to support the diagnosis and prognosis analysis of single ventricle heart disease. Compared with the traditional attention mechanism, the introduction of the channel attention mechanism can more fully capture the relationship between local and global features in the image, effectively improving the accuracy of cardiac CT image segmentation. In terms of model architecture, the designed cascaded CAUNet introduces the channel attention module in the two-stage refinement network by constructing an adaptive cascade segmentation framework; that is, the ventricular region is roughly segmented in the first stage, and the initial segmentation result is optimized in detail in the second stage. Through this two-stage segmentation strategy, the segmentation performance of the network is not only improved, but also shows strong robustness when dealing with single ventricle cases with irregular morphology. In order to verify the effectiveness of the method, an experiments on 4 real clinical data sets is conducted in this study. The experimental results show that under the premise of a small increase in computational complexity, the Dice similarity coefficient of the network proposed in this paper in the segmentation of 7 substructures of the heart, the left and right ventricles, and the pulmonary artery segmentation tasks is improved compared with the mainstream multi-organ segmentation network, proving the potential of channel attention in cardiac image segmentation applications. The obtained precise segmentation results provide accurate quantitative indicators for clinicians to assist in the diagnosis of single ventricle structural abnormalities, which has important application value.