• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

HE染色乳腺癌组织病理图像癌巢与间质分割研究

阚贤响,刘娟,屈爱平   

  1. (武汉大学计算机学院,湖北 武汉 430072)
  • 收稿日期:2016-09-19 修回日期:2016-11-24 出版日期:2017-02-25 发布日期:2017-02-25
  • 基金资助:

    江苏省自然科学基金(BK20161249)

Segmentation of tumor nests and stomata of
HE-stained breast cancer histopathological images
 

KAN Xian-xiang,LIU Juan,QU Ai-ping   

  1. (School of Computer,Wuhan University,Wuhan 430072,China)
  • Received:2016-09-19 Revised:2016-11-24 Online:2017-02-25 Published:2017-02-25

摘要:

HE染色的乳腺癌组织病理图像是分析诊断乳腺癌常用的方法,病理学家普遍认为癌巢和间质的病理形态学特征对研究乳腺癌的生物学行为有着预示作用,所以准确分割癌巢和间质显得尤为重要。对于HE染色乳腺癌组织病理图像,视癌巢和间质的分割为图像中像素点的分类问题,提取并分析特征,选取最佳特征组合,然后分类为癌巢或间质,并结合间隔采样、归一化与阈值法。实验表明,该方法能较为准确地分割出癌巢和间质,保证较高准确率和精度,在时间效率上能达到较为满意的结果。
 

关键词: HE染色, 图像分割, 分类

Abstract:

It is common to analyze and diagnose breast cancer (BC) via HE-stained BC histopathological images. Pathologists generally believe that the pathological and morphological features of tumor nests (TNs) and stroma indicate biological behavior of BC, so it becomes particularly important to accurately segment the TNs and stroma. For HE-stained BC histopathology images, we regard the segmentation as the classification of pixels in the image, extract and analyze features, choose the best combination of features, and then classify it as TNs or stroma. And procedures of sample interval, normalization and thresholding are also taken into account. Experimental results show that the proposed method can segment the TNs and stromata accurately and ensure higher accuracy and precision. What's more, it is satisfactory in terms of time efficiency.
 

Key words: HE stained, image segmentation, classification