基于深度学习的肺结节识别
Recognition of Pulmonary Nodule Based on Deep Learning
投稿时间:2018-12-05  修订日期:2019-01-24
DOI:
中文关键词: 肺结节  深度学习  小样本  迁移训练
英文关键词: Lung nodule  deep learning  small sample  transfer training
基金项目:湖北省自然科学基金资助项目(2014CFB922)
作者单位E-mail
高智勇 中南民族大学 生物医学工程学院 zhiyonggao@mail.scuec.edu.cn 
万昕 中南民族大学 生物医学工程学院  
摘要点击次数: 74
全文下载次数: 
中文摘要:
      肺结节的自动准确识别对于肺癌的诊断治疗具有重要的意义。针对肺结节识别问题,采用分支结构改进直接型VGG16的深度学习网络结构,并使用迁移训练,能在减少网络结构复杂度的同时,提高学习效率,降低方法对硬件资源的要求,增加其适用性。实验通过训练多种网络模型,对比模型特点,调整改进的网络结构,实现了对小样本的肺结节识别。在LIDC-IDRI数据集上的进行的实验结果表明,该方法在保持结构简单降低硬件资源需求的同时,能够取得较好的肺结节识别效果。
英文摘要:
      Recognition of pulmonary nodule is an important step during the diagnosis and therapy of lung cancer. In this paper, an improved VGG16 network is presented to recognize nodule. The neural network is trained by migration training and three branches are added in the network before max pooling to extract more features. Experiment results on LIDC-IDRI data show that the proposed network recognize pulmonary nodule accurately with shallow structure, which is more practical than many other network in lung cancer detection.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭