Meeting Abstract
Stromal Filters in Automated Immunostain Scoring
Kunal Patel,
1
Anthony Bui,
1,2
Greg Riedlinger,
1
and Yukako Yagi
1
1
Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA
2
Boston Children’s Hospital, 300 Longwood Avenue, Boston, MA 02115, USA
Correspondence should be addressed to Kunal Patel; kpatel1182@gmail.com
Received 2 September 2014; Accepted 2 September 2014
Copyright © 2014 Kunal Patel et al. Tis is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Introduction
KI-67 is a marker for cell proliferation which binds a nuclear
antigen and is thus a prime antibody in immunohistochem-
istry. Scoring methodologies derived from KI-67 immunos-
taining have shown promise as predictors of lethality in a
range of cancers [1]. Te most common scoring method is
the labelling index, a ratio of KI-67 positive cells to the entire
population [1].
Manual calculation of a labelling index can be a time-
consuming process for pathologists, who count cells in a
bright feld. Automated scoring algorithms and programs
have been written to generate labelling indices, but they are
nonspecifc with respect to cell type and most ofen skew
results by including stromal cells in the index. Tis problem
is particularly relevant in breast cancer, where tumors are
ofen located in a feld of fbroadipose tissue. We have adapted
algorithms for immunostain scoring and tested 2 stromal cell
fltration algorithms which remove these cells based on their
elongated nuclear morphology.
Methods
An automated scoring algorithm was adapted from Immuno-
Ratio [2] and written in Python programming language.
Within this scoring program, two methods of stromal fl-
tration were tested. Te frst fltered cells below a threshold
defned by the isoperimetric quotient of labeled cells, while
the second used a shape property known as the Hu moment
invariant. Te algorithm was applied to a dataset of breast
cancer images, and results were correlated with nonstromal
fltration and a pathologist score.
Results
Preliminary results show that algorithms with stromal fltra-
tion correlate better with pathology results than nonfltered
counterparts. Additionally, the Hu moment invariant is a bet-
ter variable for stromal cell fltration due to its nonreliance on
perimeter. Deviation of automatic scores from pathologist-
scored results is directly correlated with the amount of
stromal tissue in the image feld.
Conclusions
Stromal cell fltration is a promising technique in the devel-
opment of automatic scoring algorithms. Additional methods
should be explored to address the fltration of nonelongated
stromal nuclei. With improvements in usability, it can be
integrated into whole slide imaging systems in the near
future.
References
[1] B. K. B. Hirata, J. M. M. Oda, R. L. Guembarovski, C. B.
Ariza, C. E. C. D. Oliveira, and M. A. E. Watanabe, “Molecular
markers for breast cancer: prediction on tumor behavior,”
Disease Markers, vol. 2014, Article ID 513158, 12 pages, 2014.
[2] V. J. Tuominen, S. Ruotoistenm¨ aki, A. Viitanen, M. Jumppanen,
and J. Isola, “ImmunoRatio: a publicly available web application
for quantitative image analysis of estrogen receptor (ER),
progesterone receptor (PR), and Ki-67,” Breast Cancer Research,
vol. 12, no. 4, article R56, 2010.
Hindawi Publishing Corporation
Analytical Cellular Pathology
Volume 2014, Article ID 497426, 1 page
http://dx.doi.org/10.1155/2014/497426