基于Transformer与距离图谱的泛癌细胞核图像分割
DOI:
作者:
作者单位:

1.南京信息工程大学 自动化学院;2.人工智能学院智慧医疗研究院

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金(U1809205, 62171230,92159301,62101365, 61771249, 91959207, 81871352)


Pan Cancer Nuclear Image Segmentation Based on Transformer and Distance Map
Author:
Affiliation:

Nanjing University of Information Science and Technology

Fund Project:

National Natural Science Foundation of China(U1809205, 62171230,92159301,62101365, 61771249, 91959207, 81871352)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    摘 要: 肿瘤细胞的密度、核质比和平均尺寸等指标都对癌症的分级和预后有重要的意义。因此在计算病理学中,细胞核分割是肿瘤微环境分析的基础。此外,通过对分割结果进行统计分析,对新的肿瘤标记物的探索有重大的意义。然而,病理图像图像背景下的细胞核形态不规则,细胞核染色不均匀,且细胞核边缘之间存在粘连的问题。现有的深度学习算法在细胞核主体分割正确的情况下,边缘的分割错误不会对总体的损失造成太大的影响,因此粘连的细胞核很容易被当作同一个分割目标。为了解决细胞核重叠问题,本文提出一种基于Transformer与距离图谱的分割模型,将Transformer中的核心多头自注意力机制与距离图谱引导算法相结合,重视细胞核内部,弱化细胞核边界,提升对图像局部和全局上下文信息的学习能力。本文方法在两个公开数据集上的平均Dice系数为0.7979、精度为0.7561、AJI系数为0.6672、Hausdorff距离为10.11。实验结果表明,相较其它分割算法,本文方法的性能更好,能够有效提高细胞核的分割精度,同时较好地解决了细胞核之间的粘连问题。

    Abstract:

    Abstract: Indices such as tumor cell density, nucleocytoplasmic ratio, and average size have important implications for cancer grading and prognosis. Therefore, nuclear segmentation is the basis of tumor microenvironment analysis in computational pathology. In addition, the exploration of new tumor markers is of great significance through statistical analysis of segmentation results. However, the morphology of nuclei in the background of pathological images is irregular, the staining of nuclei is uneven, and there is a problem with adhesion between the edges of nuclei. In the existing deep learning algorithm, when the main body of the nucleus is correctly segmented, the segmentation error of the edge will not have much impact on the overall loss, so the adhering nucleus can easily be regarded as the same segmentation target. In order to solve the nuclear overlapping problems, a new segmentation algorithm based on the Transformer and Distance map, TDM-Net, is proposed, which integrates the core of multi-head self-attention mechanism in Transformer with contextual information to fully explore the proximity relationship and enhances the learning ability of image details by introducing distance map to emphasize the interior of nuclei and weaken the boundary of nuclei. The algorithm's dice coefficient, precision, aggregated jaccard index and hausdorff distance are 0.7979, 0.7561, 0.6672, and 10.11, respectively. The results show that TDM-Net performs better than other segmentation algorithms, effectively improving nuclei segmentation accuracy and solving overlapping problems with different nuclei.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2022-05-21
  • 最后修改日期:2022-07-15
  • 录用日期:2022-07-16
  • 在线发布日期:
  • 出版日期:

地址:江苏南京,宁六路219号,南京信息工程大学    邮编:210044

联系电话:025-58731025    E-mail:nxdxb@nuist.edu.cn

南京信息工程大学学报 ® 2022 版权所有  技术支持:北京勤云科技发展有限公司