Sunteți pe pagina 1din 4

2017 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS)

Segmentation of thermal infrared breast images using


K-Means, FCM and EM algorithms
for breast cancer detection
R.Meena Prakash K. Bhuvaneshwari, M. Divya, K. Jamuna Sri,
Associate Professor, Department of Electronics and A.Sulaiha Begum,
Communication Engineering Department of Electronics and Communication Engineering
V.P.M.M.Engineering College for Women V.P.M.M.Engineering College for Women
Krishnankoil, India Krishnankoil, India
meenaprakash73@gmail.com

analysis and Support Vector Machine (SVM) classifier [3].


Abstract— Breast cancer is the most commonly occurring Mahmoudzadeh et al proposed an Extended Hidden Markov
cancer among women leading to deaths. Detection of tumors at Model for segmentation of breast thermogram [4]. Color
early stages improves the survival rate. Infrared thermograph segmentation of breast infrared images using Gaussian Mixture
images of breast provide information based on temperature Model (GMM) where the model parameters are estimated
changes in the breast. This information is more supportive in
using Expectation Maximization (EM) algorithm is proposed
detecting cancer at its early stage. In this paper, three
segmentation techniques, K-Means, Fuzzy C- Means (FCM) and [5]. Harrabi and Braiek proposed color image segmentation
Gaussian Mixture Model - Expectation Maximization (GMM- based on modified Fuzzy C – Means algorithm and different
EM) are employed to segment the IR breast images and color spaces [6]. Two color segmentation techniques, K-Means
compared. The method is applied to classify the malignant and and Fuzzy C – Means are modeled and compared for color
benign cancer tissues. segmentation of IR breast images [7]. A thermographic image
analysis method based on Independent Component Analysis
Index Terms— FCM, EM, K-Means, Colour Segmentation, (ICA) for automatic detection of high tumor risk areas is
breast images. proposed [8]. Etehadtavakol et al employed bispectral
I. INTRODUCTION invariant features for classification of benign and malignant
classes of images [9]. Acharya et al proposed automatic
Breast cancer is the second significant cause of death in identification of breast cancer using GLCM feature extraction
women. Early detection of breast cancers decreases the cancer from thermograph images [10]. A method of breast cancer
deaths among women. The most commonly available methods detection in rotational thermography images using texture
for detecting the breast cancer include Mammography, Breast features extraction is proposed [11]. A method of segmentation
Ultrasound, Magnetic Resonance Imaging (MRI), Breast of thermogram images for the identification of breast cancer
Thermography and Positron Emission Tomography (PET) etc., using projection profile approach is proposed [12]. Automated
Breast thermography is the non-invasive, non-radiating, fast, segmentation and detection methods of breast cancer in
painless and low cost imaging method for early detection of mammograms are proposed [13, 14].
breast cancer. It is an effective method of imaging for all In this paper, segmentation of infrared breast images using
women irrespective of ages, sizes and types of breast. The three algorithms, K-Means, Fuzzy C- Means (FCM) and
method is also applicable to monitor the breast condition after Gaussian Mixture Model (GMM) followed by colour analysis
surgery. is proposed to detect breast cancer in its early stage.
The breast thermography uses infrared camera to generate
high resolution images of the temperature changes in the II. PROPOSED METHOD
breast. Precancerous and cancerous masses are highly
The overview of the proposed method is shown in figure
proliferating tissues and they need large amount of nutrients.
1. The input infrared breast image in RGB color format is first
Hence, they derive new blood vessels which results in increase
converted into L*a*b* color space since the color
in surface temperature. Thermography can detect the indication
segmentation has the advantages of detecting the abnormality
of cancer earlier than mammography. better and determining the extent of malignant cells. In L*a*b*
Shahari and Wakankar employed the K-Means color space, L* denotes the luminosity layer, a* denotes the
segmentation followed by color analysis for breast cancer chromaticity layer where the color falls along the red green
detection in infrared images [1]. Hankare et al performed color axis and b* denotes the chromaticity layer where the color falls
based abnormality separation in thermograph images of breast along the blue yellow axis. After color space conversion, three
using K-Means segmentation [2]. Wakankar and Suresh different segmentation algorithms K-Means, Fuzzy C-Means
proposed automatic diagnosis of breast cancer using color

‹,(((
2017 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS)

and Gaussian Mixture Model – Expectation Maximization Step 4: Repeat steps 2 and 3 till convergence criterion given by
algorithm are applied. The different clusters are labeled and the equation (3) is met.
hottest regions are identified. Based on that, the images are ‖ − −1 ‖ < 0.00001 (3)
classified into benign and malignant classes. where represents the iteration.
C. GMM – EM Algorithm
Input the IR thermal breast image in Given a Gaussian Mixture Model, the likelihood function
RGB format should be maximized with respect to the parameters – Means
and Variances of GMM. The parameters are estimated using
the EM algorithm and the steps are given below.
Convert the color space into Step 1: Initialize the number of classes, means, variances and
L*a*b* space the mixing coefficients.
Step 2: Expectation Step: Compute the posterior probabilities
with the current parameter values using equation 4.
( / ,Σ )
Perform K-Means / Fuzzy C – ( )= (4)
∑ =1 ( / ,Σ )
Means / GMM-EM Segmentation
where (. ) represents the Gaussian Mixture Model.
Step 3: Maximization Step: The parameters – Means, variances
and mixing coefficients are computed using the current
Label the regions into clusters and posterior probabilities using equations (5), (6) and (7)
identify the hottest region respectively.
∑ ( )
= ∑ ( )
(5)
Figure 1 – Proposed Method ∑ − −
Σ = ∑ ( )
(6)
1
A. K-Means Algorithm = ∑ ( ) (7)
The K-Means algorithm segments the input data points Step 4: Evaluate the log likelihood
into K- clusters. The steps in the K-Means algorithm are given ln ( / , Σ, ) = ∑ ∑ =1 ( / ,Σ ) (8)
below. If the convergence criterion is not satisfied, return to step 2.
Step 1: Initialize the number of clusters and the mean values of
the clusters. III. E XPERIMENTAL RESULTS
Step 2: Calculate the distance of the image intensities from the In the proposed work, infrared images of four different
means of all the clusters. cases – one normal case and three cancer cases [15, 16] are
Step 3: Cluster and label the image points based on the segmented and the work is implemented using MATLAB. The
calculated distance. images are segmented using three algorithms – K-Means, FCM
Step 4: Compute the new mean values for each cluster. and GMM-EM.
Step 5: Repeat steps 2, 3 and 4 until the labels of the clusters For FCM and EM segmentation, the color channel a* is
stabilize. considered. In figure 2 and 4, fibrocystic case breast images are
B. Fuzzy C-Means (FCM) Algorithm segmented and the results of K-Means, EM and FCM
segmentation are shown. The hottest regions are indicated in
The FCM segmentation is an unsupervised technique and figures 2 and 4 (d), (h) and (l). Segmentation of normal breast
the steps in the algorithm are given below. image is shown in figure 3 and malignant inflammatory cancer
Step 1: Initialize the number of clusters, cluster centers and case in figure 5.
construct the vector = { 1 , 2 , . . } with all the gray level
intensities of the image.
Step 2: Calculate the membership function for each gray level
intensity as given in equation (1).
1
= 2 (1)
−1
∑ =1

where represents the number of clusters, represents the (a)


degree of fuzziness and represents the distance from the
cluster center.
Step 3: Calculate the new cluster center using equation (2)
∑ =1
= ∑ =1
(2)
(b) (c) (d) (e)
2017 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS)

(f) (g) (h) (i)


(j) (k) (l) (m)
Figure 4 – Segmentation of IR thermal breast image –
Fibrocystic changes in left breast
(a) Original image (b)-(e) – Segmentation by K–Means
(j) (k) (l) (m) (f)-(i) – Segmentation by EM (j)-(m) – Segmentation by FCM
Figure 2 – Segmentation of IR thermal breast image –
Fibrocystic Case
(a) Original image (b)-(e) Segmentation by K–Means
(f)-(i) Segmentation by EM (j)-(m) Segmentation by FCM

(a)

(a)

(b) (c) (d) (e)

(b) (c) (d) (d) (f)

(g) (h) (i) (j) (k)


(f) (g) (h) (i)
Figure 5 – Segmentation of IR thermal breast image which
shows cancer case with an increase in heat and blood vessel
activity in left breast when compared to right
(l) (m) (n) (o) (p) (a) – Original image (b)-(i) – Segmentation by K–Means
Figure 3 – Segmentation of IR thermal breast image – Normal The results of implementation of K-Means segmentation of
Case (a) Original image (b)-(f) Segmentation by K–Means IR breast image of cancer case is shown in figure 5 where it
(g)-(k) Segmentation by EM (l)-(p) Segmentation by FCM shows an increase in heat and blood vessel activity in the left
breast when compared to right. The hottest region is indicated
in figure 5 (h). The qualitative analysis indicates that the FCM
segmentation gives good accuracy and indication of the
disease. The drawback of K-Means segmentation is that it
results in empty clusters in certain cases.

(a) IV. CONCLUSION


In this paper, a method to segment the IR breast images
using three segmentation techniques, K-Means, Fuzzy C-
Means (FCM) and Gaussian Mixture Model (GMM) –
Expectation Maximization (EM) is implemented. The
conversion of colour space from RGB to L*a*b* is done to
enhance the colour analysis for classification of the images into
(b) (c) (d) (e)
benign and malignant cases.
REFERENCES

[1] S. Shahari and A. Wakankar, “Color analysis of thermograms for


breast cancer detection,” 2015 International Conference on
(f) (g) (h) (i) Industrial Instrumentation and Control (ICIC), 2015.
2017 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS)
[2] P. Hankare, K. Shah, Deepthy Nair and Divya Nair, “Breast
Cancer Detection Using Thermography,” International Research
Journal of Engineering and Technology, pp.1061-1064, 2016.
[3] A. T. Wakankar and G. R. Suresh, “Automatic Diagnosis of
Breast Cancer using Thermographic Color Analysis and SVM
Classifier,” Advances in Intelligent Systems and Computing
Intelligent Systems Technologies and Applications 2016, pp. 21–
32, 2016.
[4] E. Mahmoudzadeh, M. Montazeri, M. Zekri, and S. Sadri,
“Extended hidden Markov model for optimized segmentation of
breast thermography images,” Infrared Physics & Technology,
vol. 72, pp. 19–28, 2015.
[5] S. Kermani, N. Samadzadehaghdam, and M. Etehadtavakol,
“Automatic color segmentation of breast infrared images using a
Gaussian mixture model,” Optik - International Journal for Light
and Electron Optics, vol. 126, no. 21, pp. 3288–3294, 2015.
[6] R. Harrabi and E. B. Braiek, “Color image segmentation using a
modified Fuzzy C-Means technique and different color spaces:
Application in the breast cancer cells images,” 2014 1st
International Conference on Advanced Technologies for Signal
and Image Processing (ATSIP), 2014.
[7] M. Etehadtavakol, S. Sadri, and E. Y. K. Ng, “Application of K-
and Fuzzy c-Means for Color Segmentation of Thermal Infrared
Breast Images,” Journal of Medical Systems, vol. 34, no. 1, pp.
35–42, 2008.
[8] L. Boquete, S. Ortega, J. M. Miguel-Jiménez, J. M. Rodríguez-
Ascariz, and R. Blanco, “Automated Detection of Breast Cancer
in Thermal Infrared Images, Based on Independent Component
Analysis,”Journal of Medical Systems, vol. 36, no. 1, pp. 103–
111, Oct. 2010.
[9] M. Etehadtavakol, V. Chandran, E. Ng, and R. Kafieh, “Breast
cancer detection from thermal images using bispectral invariant
features,” International Journal of Thermal Sciences, vol. 69, pp.
21–36, 2013.
[10] U. Acharya, E. Ng, S. Subbhuraam, S. Suri, “Automatic
Identification of Breast Cancer Using Texture Features from
Infrared Thermograms,”Diagnostic and Therapeutic Applications
of Breast Imaging, pp. 399–418.
[11]S. V. Francis, M. Sasikala, G. B. Bharathi, and S. D. Jaipurkar,
“Breast cancer detection in rotational thermography images using
texture features,” Infrared Physics & Technology, vol. 67, pp.
490–496, 2014.
[12]D. Dayakshini, S. Kamath, K. Prasad, and K. V. Rajagopal,
“Segmentation of Breast Thermogram Images for the Detection
of Breast Cancer – A Projection Profile Approach,” Journal of
Image and Graphics, vol. 3, no. 1, 2015.
[13]D. C. Pereira, R. P. Ramos, and M. Z. D. Nascimento,
“Segmentation and detection of breast cancer in mammograms
combining wavelet analysis and genetic algorithm,” Computer
Methods and Programs in Biomedicine, vol. 114, no. 1, pp. 88–
101, 2014.
[14]A. R. Domínguez and A. K. Nandi, “Toward breast cancer
diagnosis based on automated segmentation of masses in
mammograms,” Pattern Recognition, vol. 42, no. 6, pp. 1138–
1148, 2009.
[15]http://www.thermologyonline.org/Breast/breast_thermography_
what.htm
[16]http://aathermography.com/
[17]http://www.breastthermography.com/case_studies.htm

S-ar putea să vă placă și