Image Segmentation of Nucleus Breast Cancer using Digital Image

Processing

Ana Yulianti

, Ause Labellapansa

, Evizal Abdul Kadir

, Mohana Sundaram

, and Mahmod Othman

Department of Informatics Engineering, Universitas Islam Riau, Pekanbaru, Indonesia

Department of Fundamental & Applied Sciences, Universiti Teknologi Petronas, Perak Darul Ridzuan, Malaysia

Keywords:

IHC Breast Cancer, ER/PR Receptor, Image Processing

Abstract:

One of examination methods of breast cancer cells is using Immunohistochemistry (IHC). IHC is used to

determine the status of Estrogen Receptor (ER) and/or Progesterone Receptor (PR). The bonding reaction

occurring between the cell and the painting results in the color of the nucleus cell being blue which signiﬁes

the negative and brown ER/PR hormone for positive ER/PR. The given hormonal therapy will be effective to

breast cancer patients if they have positive ER/PR receptors. Up to now the Anatomy Pathology specialist

calculatses the percentage of positive cells that have been marked semiquantitatively. This is time-consuming,

costly, subjective and tedious, thereby impacting the length of time required in determining appropriate therapy

for breast cancer patients. This study analyze the image of IHC breast cancer to determine the assessment of

ER/PR hormone receptor using image processing. The use of kernels of different sizes shows differences in

the results of cell segmentation in connective tissue. The use of 3×3 and 1×1 kernels has indeed succeeded in

removing cells in the connective tissue, but not all cells in the connective tissue can be identiﬁed. If this step

has been completed, then the next process until cell count can be done.

1 INTRODUCTION

Breast cancer is a dangerous disease that occurs due

to the uncontrolled cells growth. One of examination

methods of breast cancer cells is using Immunohisto-

chemistry (IHC). IHC is used to determine the status

of Estrogen Receptor (ER) and/or Progesterone Re-

ceptor (PR). The IHC technique is performed by ap-

plying Hematoxylin and Diaminobenzidine and ob-

serving the antibody presence bonds by microscope

based on the observation by Pathologyst. The bond-

ing reaction occurring between the cell and the paint-

ing results in the color of the nucleus cell being blue

which signiﬁes the negative and brown ER/PR hor-

mone for positive ER/PR. The given hormonal ther-

apy will be effective to breast cancer patients if they

have positive ER/PR receptors. IHC image Positive

estrogen receptors and negative show in the Figure 1.

Up to now the Anatomy Pathology specialist cal-

culatses the percentage of positive cells that have been

marked semiquantitatively. This is time-consuming,

costly, subjective and tedious (Limsiroratana and

Boonyaphiphat, 2009; Estrogen, ), thereby impacting

the length of time required in determining appropriate

therapy for breast cancer patients. This study will an-

(a) (b)

Figure 1: IHC image (a) Positive estrogen receptors (b)

Negative estrogen receptors

alyze the image of IHC breast cancer to determine the

assessment of ER / PR hormone receptor using digital

image processing which is expected to help doctors to

determine whether the breast cancer patients require

hormonal therapy or not.

(Kostopoulos et al., 2007; Calhoun et al., 2019)

provides a positive estrogen receptor assessment by

analyzing IHC images using a color texture feature.

Assessment of positive ER receptor status with com-

puter method is done through 2 stages, ie. stage I

of segmentation of nucleus using Otsu’s global im-

age threshold method and morphology operation and

stage II classiﬁcation of nucleus based on brown

and blue color using feature selection and K-Nearest

Yulianti, A., Labellapansa, A., Kadir, E., Sundaram, M. and Othman, M.

Image Segmentation of Nucleus Breast Cancer using Digital Image Processing.

DOI: 10.5220/0009105900640067

In Proceedings of the Second International Conference on Science, Engineering and Technology (ICoSET 2019), pages 64-67

ISBN: 978-989-758-463-3

Neighbors weighted voted (KNN-WV).

(Yulianti et al., 2014; Akbari et al., 2011) seg-

mented the immunohistochemical image of estrogen

receptor to breast cancer using watershed marker.

(Labellapansa et al., 2016) conducted a similar study

but using the IHC HER2 method for scores of 1+ and

3+ and the classiﬁcation can be done correctly 100%

for scores of 3+ and 65% for scores of 1+.

This study was able to indicate the status of ER /

PR and remove the stacked cell area however the con-

nective tissue cell that is not a nucleus cell counts as

ER / PR cell as shown in Figure 2. Our research will

make image improvements by removing connective

tissue that is not a nucleus cell which hopefully will

be able to calculate the number of ER / PR cells in

more detail.

Figure 2: IHC image positive estrogen receptors

2 RESEARCH METHOD

The ﬁrst step is done by acquiring positive / negative

images of ER / PR in the lab Medicine Faculty of Ga-

jah Mada University. The phases of pre processing

the imagery are done by using the median ﬁltering

method. The clean image of the noise, will enter the

segmentation stage to separate the blue area (negative

cell ER / PR) and brown area (positive cell ER / PR)

using colour deconvolution. The Color deconvolution

method can read the colors of each channel Red Green

Blue (RGB)(Ruifrok et al., 2001). Watershed is used

to separate the stacked cell area using color decon-

volution. The next step which is the most important

contribution in this study is to identify the connective

tissue that is not a cell. The shape feature will be used

to remove this connective tissue area. The next step

is to calculate the portion of positive cell and nega-

tive cell so that can be identiﬁed whether the image is

positive ER / PR or negative. Flow Chart of Research

Activities as shown in Figure 3.

Figure 3: Flow Chart of Research Activities

3 RESULT AND DISCUSSION

The phases of nucleus IHC breast cancer image seg-

mentation are shown in Figure 4. The input image

(a) is pre-processed using the median ﬁlter (b) then

color segmentation is done using color deconvolution

so that the image of channel 1 H (c) and channel 2

DAB Positive (d) The next step is to separate the ac-

cumulated cells in the H image and the positive DAB

image by using watershed marker segmentation. Fig-

ures 4 (e) and (f) are the results of the segmentation

so that it is expected that the number of cells can be

calculated.

Morphological reconstruction was carried out af-

ter the process in Figure 4 was completed. Mor-

phological reconstruction is a morphological transfor-

mation involving two images and one structural ele-

ment. The ﬁrst image is the start point of transfor-

mation, commonly referred to as the marker and the

second image as a constraint, commonly referred to as

a mask. The process of morphological transformation

is based on the concept of pixel neighbors using struc-

tural elements (Gonzalez et al., 2002). Pixel neighbor

operation is an image processing operation to get the

value of a pixel that involves neighboring pixel values

and is mostly used for form analysis (Kadir, 2017).

The use of kernels of different sizes shows differ-

ences in the results of cell segmentation in connective

tissue. This study uses a kernel size of 3x3 and 1x1.

Image Segmentation of Nucleus Breast Cancer using Digital Image Processing

(a)

(b)

(c)

(d)

(e)

(f)

Figure 4: The Stage of IHC Nucleus Image Segmenta-

tion Stage of Breast Cancer (a) Image of estrogen recep-

tor (b) Image Resulted by Median ﬁltering (c) Image H Re-

sulted by colour Deconvolution (d) Image DAB Resulted by

Colour Deconvolution (e) Image H Resulted by Watershed

(f) Image DAB Resulted by Watershed

(a) (b)

Figure 5: Image of the result of using (a) 3x3 Kernel Size

(b) 1x1 kernel size

Figure 5 is the result of cell segmentation in connec-

tive tissue using a 3x3 and 1x1 disk kernel. It is seen

that cells in the connective tissue are still counted as

many cancer cells while using a 1x1-sized kernel seen

in cells in the connective tissue there are not many

counts.

Based on the results seen in ﬁgure 5, the use of

3x3 and 1x1 kernels has indeed succeeded in remov-

ing cells in the connective tissue, but not all cells in

the connective tissue can be identiﬁed. This research

will be continued by using other methods to remove

cells in connective tissue. If this step has been com-

pleted, then the next process until cell count can be

done.

4 CONCLUSIONS

From the steps that have been done above, some re-

sults are obtained, namely stages of digital image pro-

cessing to read IHC breast cancer images to obtain

H cell counts and positive DAB cell numbers begin-

ning with the pre-processing process using Median

Filtering, then proceed with colour segmentation us-

ing Colour Deconvolution to obtain IHC H images

and positive DAB IHC images and followed by cell

segmentation using Watershed Markers. The use of

3x3 and 1x1 kernels has indeed succeeded in remov-

ing cells in the connective tissue, but not all cells in

the connective tissue can be identiﬁed. This research

will be continued by using other methods to remove

cells in connective tissue. If this step has been com-

pleted, then the next process until cell count can be

done.

ACKNOWLEDGEMENTS

High appreciation should be given to Universitas Is-

lam Riau (UIR) and Universiti Teknologi Petronas

(UTP) for their support in matching grant of this re-

search work.

REFERENCES

Akbari, H., Uto, K., Kosugi, Y., Kojima, K., and Tanaka, N.

(2011). Cancer detection using infrared hyperspectral

imaging. Cancer science, 102(4):852–857.

Calhoun, M. E., Chowdhury, S., and Goldberg, I. (2019).

Medical analytics system. US Patent 10,255,997.

Estrogen, S. C. I. R. Kanker payudara menggunakan marker

watershed.

Gonzalez, R. C., Woods, R. E., et al. (2002). Digital image

processing.

Kadir, A. (2017). Teori dan aplikasi pengolahan citra.

Kostopoulos, S., Cavouras, D., Daskalakis, A., Bougioukos,

P., Georgiadis, P., Kagadis, G. C., Kalatzis, I., Rava-

zoula, P., and Nikiforidis, G. (2007). Colour-texture

based image analysis method for assessing the hor-

mone receptors status in breast tissue sections. In 2007

29th Annual International Conference of the IEEE

Engineering in Medicine and Biology Society, pages

4985–4988. IEEE.

Labellapansa, A., Muhimmah, I., and Indrayanti (2016).

Segmentation of breast cancer cells positive 1+ and 3+

immunohistochemistry. In AIP Conference Proceed-

ings, volume 1718, page 110002. AIP Publishing.

Limsiroratana, S. and Boonyaphiphat, P. (2009). Computer-

aided system for microscopic images: application

to breast cancer nuclei counting. INTERNATIONAL

JOURNAL OF APPLIED, 2(1):69.

ICoSET 2019 - The Second International Conference on Science, Engineering and Technology

Ruifrok, A. C., Johnston, D. A., et al. (2001). Quantiﬁ-

cation of histochemical staining by color deconvolu-

tion. Analytical and quantitative cytology and histol-

ogy, 23(4):291–299.

Yulianti, A., Muhimmah, I., and Indrayanti, I. (2014).

Segmentasi Citra Imunohistokimia Reseptor Estro-

gen Kanker Payudara menggunakan Marker Water-

shed. Seminar Nasional Informatika Medis (SNIMed),

0(5):1–10.

Image Segmentation of Nucleus Breast Cancer using Digital Image Processing