陆雪松,闫书豪.基于迭代卷积神经网络的肝脏MRI图像分割[J].中南民族大学学报自然科学版,2022,41(3):319-325
基于迭代卷积神经网络的肝脏MRI图像分割
Iterative convolutional neural network for liver MRI segmentation
  
DOI:10.12130/znmdzk.20220310
中文关键词: 肝脏分割  腹部磁共振  迭代卷积神经网络
英文关键词: liver segmentation  abdominal magnetic resonance imaging  iterative convolutional neural network
基金项目:湖北省自然科学基金资助项目(2016CFB489)
作者单位
陆雪松 中南民族大学 生物医学工程学院武汉 430074 
闫书豪 中南民族大学 生物医学工程学院武汉 430074 
摘要点击次数: 231
全文下载次数: 233
中文摘要:
      从医学图像中快速准确地提取肝脏组织是精准诊疗的基础.针对腹部MRI中呈现出浸润现象、与相邻器官灰度值相似、边界较为模糊的问题,提出了一种基于深度卷积编解码的迭代网络结构.它将前次分割输出的概率图引入到网络浅层,与来自编码器的低水平特征图融合,弥补下采样时丢失的细节信息,迭代地完成网络参数更新.通过在ISBI 2019 liver-chaos挑战数据集上的验证实验,该方法的分割精度较传统U-Net有明显提高,能够更好地服务于临床工程.
英文摘要:
      Rapid and accurate extraction of liver tissue from medical images is the basis of precision diagnosis and therapy. To tackle the problems that abdominal MRI presents infiltration phenomenon, adjacent organs with similar intensities, and the ambiguous boundary, an iterative network architecture based on deep convolutional encoder-decoder is proposed. In order to compensate for the loss of detailed information in down-sampling, the probability map from previous segmentation is introduced to the shallow layer of the network, which is fused with the low-level feature map. And then, network parameters are updated iteratively. Experiments were performed on dataset from ISBI 2019 liver-chaos challenge. The results show that segmentation accuracy of the proposed method is significant better than that of traditional U-Net, which can be better for clinical practice.
查看全文   查看/发表评论  下载PDF阅读器
关闭