唐奇伶,方全,夏先富,杨济榕.基于堆栈稀疏自编码与整体嵌套的乳腺病理图像细胞识别[J].中南民族大学学报自然科学版,2019,(3):397-403
基于堆栈稀疏自编码与整体嵌套的乳腺病理图像细胞识别
Breast pathology image cell identification based on stacked sparse autoencoder and holistically-nested structure
  
DOI:10.12130/znmdzk.20190315
中文关键词: 细胞识别  堆栈稀疏自编码  整体嵌套  病理图像
英文关键词: cell identification  stacked sparse autoencoder  holistically-nested  pathological images
基金项目:中央高校基本科研业务费专项资金 (CZZ18005)
作者单位
唐奇伶,方全,夏先富,杨济榕* 中南民族大学 生物医学工程学院武汉 430074 
摘要点击次数: 62
全文下载次数: 68
中文摘要:
      为解决病理图像的快速分析和利用细胞自动识别的问题,提出了一种基于堆栈稀疏自编码与整体嵌套的细胞识别的方法,快速、高效、准确地识别了高分辨率病理组织图像中的细胞.运用堆栈自编码对训练集中的细胞样本块和非细胞样本块提取高级特征,并运用特征训练Softmax分类器,加入融合层进行融合,用于细胞的自动识别;在训练过程中,还引入了一种新的整体嵌套结构.结果表明:该算法能有效识别出细胞,较其他自编码相比,具有更高的准确率、召回率和综合评价指标.
英文摘要:
      To solve the problem of rapid analysis of pathological images and automatic identification of cells, a method based on stacked sparse autoencoder and holistically-nested cell detection was proposed, which can identify cells in high-resolution tissue images quickly, efficiently and accurately. The stack autoencoder was used to extract the high-level features of the fast and non-cell samples in the training set. Then, the Softmax classifier was trained by the feature, and the fusion layer was added for fusion, to identify the cell automatically. Then it is used for the automatic identification of the cells. In the training process, a new holistically-nested structure was introduced. The experimental results showed that the algorithm could identify the cells effectively, with higher accuracy, recall rate and more comprehensive evaluation index compared with other autoencoder.
查看全文   查看/发表评论  下载PDF阅读器
关闭