朱勇,谢勤岚.3D SE-ResNet:一种从CT图像中自动分割COVID-19肺部感染模型[J].中南民族大学学报自然科学版,2022,41(2):200-207
3D SE-ResNet:一种从CT图像中自动分割COVID-19肺部感染模型
3D SE-ResNet:an automatic segmentation of COVID-19 lung infection model from CT images
  
DOI:10.12130/znmdzk.20220211
中文关键词: 新型冠状病毒  3D分割  计算机断层  深度学习  人工智能
英文关键词: COVID-19  3D segmentation  computed tomography  deep learning  artificial intelligence
基金项目:湖北省自然科学基金资助项目(2016CFB489)
作者单位
朱勇 中南民族大学 生物医学工程学院武汉 430074 
谢勤岚 中南民族大学 生物医学工程学院武汉 430074 
摘要点击次数: 543
全文下载次数: 395
中文摘要:
      受新型冠状病毒感染的肺部,在胸部CT图像中表现出非常明显的病理特征,为此开发了一种自动分割模型能够帮助医生进行有效的诊断和分析.基于3D U-Net分割框架,提出了一种结合挤压与激励模块和注意力机制结合的3D SE-ResNet深度学习模型.采用包含200张COVID-19 3D CT图像数据集并通过大量的数据增强后对模型进行了训练和评估.实验结果表明:所提出的模型Dice指标达到了87.00%,与基于3D ResNet和3D U-Net网络模型实验相比,分别提升2.85%和1.03%.可以看出,3D SE-ResNet网络模型能在冠状病毒感染区域分割实现较高精度.
英文摘要:
      The lung infected by novel coronavirus showed obvious pathological features in chest CT images, and the development of an automatic segmentation model can help physicians to make effective diagnosis and analysis.Based on 3D U-Net segmentation framework, a 3D SE-ResNet deep learning model combining extrusion and excitation modules and attention mechanism was proposed.The model was trained and evaluated using a data set of 200 COVID-19 3D CT images and enhanced with a large number of data.The experimental results show that the Dice index of the proposed model reaches 87.00%, which increases by 2.85% and 1.03% respectively compared with the experiment based on 3D ResNet and 3D U-NET network model.It can be seen that 3D SE-ResNet network model can achieve high accuracy in coronavirus infected region segmentation.
查看全文   查看/发表评论  下载PDF阅读器
关闭