Friday, March 22nd, 2019


A full text version of this article is available.
To access article obtain online access here or login
Detection of Ki67 Expression by Analyzing Texture of Hematoxylin-and-Eosin–Stained Images, the Effectiveness of Signal Intensity, and Co-occurrence Matrix Features
Authors:  Fumikazu Kimura, Ph.D., C.M.I.A.C., Masahiro Ishikawa, Ph.D., Sercan Taha Ahi, Ph.D., Chamidu Atpelage, Ph.D., Yuri Murakami, Ph.D., Jun Watanabe, M.D., Ph.D., F.I.A.C., Hiroshi Nagahashi, Ph.D., and Masahiro Yamaguchi, Ph.D.
  Objective: To evaluate whether it is possible to determine the nuclei of immunohistochemical expression for Ki-67 from hematoxylin-and-eosin–stained (HE) specimens. Although in the pathological diagnosis we have confirmed the proliferative situation of a tumor, in general this only confirms the proliferative state of one part of the tumor and it is not possible to ascertain the proliferative state of the whole tumor. If the proliferation ability of the whole tumor can be ascertained, a more accurate treatment and prognosis presumption will be possible.
Study Design:
The HE stain and immunohistochemical reaction for Ki-67 was performed with the same specimen in endometrial adenocarcinoma of the uterine corpus. Ki-67 negative and positive nuclei were selected from HE digital images. Texture and morphological analysis were calculated from selected nuclei in HE images.
Texture and morphological features were significantly different between Ki-67 negative and positive nuclei. When a discriminant analysis was performed with a linear support vector machine using these features, the accuracy for separating the Ki-67 negative and positive nuclei was 85.6%.
A discriminant analysis using features obtained through texture and morphological analysis are thought to be useful in identifying nuclear expression of Ki-67 in HE specimens.
Keywords:  co-occurrence matrix, Ki-67, signal intensity, support vector machine, texture analysis
  Acrobat Reader 7.0 is recommended to properly view and print the article.
Reader can be downloaded from