Development of a Medical Image Segmentation Algorithm based on

Fuzzy C-Means Clustering

EL FAHSSI KHALID1, OUNASSER SAIDA2, MOHAMED TAJ BENNANI3,

ABENAOU ABDENBI2

1Department Computer Science, LISAC,

Faculty of Science Dhar El Mahraz, Sidi Mohamed Ben Abdellah University of Fez,

MOROCCO

2Applied Mathematics and Intelligent Systems Engineering (MAISI),

Ibn Zohr University, ENSA-Agadir,

MOROCCO

3LPAIS Laboratory,

Computing Science Department, Faculty of Sciences Dhar El Mahraz Fez,

Sidi Mohamed Ben Abdellah University,

Fez,

MOROCCO

Abstract: - Breast mass segmentation in mammography plays a very important role in computer-aided

diagnosis (CAD) systems. In this article, we propose a mammography image segmentation method based on a

combined approach. The fuzzy clustering method and thresholding segmentation. Subsequently, we use the

wavelet transform and the Canny filter for edge detection.

Key-Words: - Images, Processing, Breast cancer, Segmentation, Mammography, CAD.

Received: July 29, 2023. Revised: October 28, 2023. Accepted: December 15, 2023. Published: December 31, 2023.

1 Introduction

Modern methods of medical diagnosis and

biomedical research largely rely on the analysis of

images obtained through various technical means

(optical and electron microscopes, X-ray and

thermographic devices, tomography, etc.). However,

effectively addressing diagnostic and scientific

challenges when using visual information

necessitates knowledge of specific methods for

image acquisition, registration, digital processing,

and analysis. This becomes particularly evident

when utilizing new types of information systems

that solve the problem of extracting hidden

information for diagnosis (e.g., CT scanners, laser

confocal microscopes, ultrasound diagnostic

devices, etc.).

Magnetic Resonance Imaging (MRI) is a

powerful method in medical diagnosis as it enables

the extraction of valuable information about the

patient's internal organs and tissues. MRI is widely

used to monitor disease dynamics such as breast

cancer, Alzheimer's disease, brain tumors, and more.

Visual inspection of these images allows specialists

to detect the emergence of certain anomalies.

However, the massive number of medical images

stored in the database makes visual analysis

challenging, and in some cases, it may not lead to a

better diagnostic outcome. Indeed, diagnostic tools

and digital image processing methods significantly

influence the diagnostic result.

One of the current trends in the development of

medical informatics is digital image processing,

including image quality enhancement, restoration of

damaged images, and recognition of individual

elements. Diagnosis and identification of

pathological processes are among the most

important tasks in the processing and analysis of

medical images. Early diagnosis of the mentioned

diseases can reduce mortality rates among patients.

To automate the process of MRI image analysis and

improve diagnostic outcomes, it is necessary to

develop new algorithms capable of addressing

problems such as image classification, contour

detection, and image segmentation.

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2023.19.23

El Fahssi Khalid, Ounasser Saida,

Mohamed Taj Bennani, Abenaou Abdenbi

E-ISSN: 2224-3488

215

Volume 19, 2023

2 Segmentation Methods

Numerous studies have been conducted in the field

of medical image processing, [1], [2], such as the

utilization of Magnetic Resonance Imaging [3], the

extraction of white brain tissue regions and single-

cell detection, [4] and topological visualization of

human brain proliferation via MRI, [5]. However,

there is no single method that would provide an

acceptable analysis for any image. For the diagnosis

of brain tumors, for example, many researchers

employ a wide range of techniques based on MRI

image segmentation. Regardless of the approach

used, the problem of MRI image segmentation

remains one of the fundamental challenges in digital

image analysis and processing.

We can categorize segmentation methods into three

main classes:

Region-based segmentation [6], edge-based

segmentation [7], and pixel-based classification or

thresholding segmentation, [8].

Methods within the first class [9], [10], [11],

[12], search for sets of pixels that exhibit a certain

degree of similarity. These techniques reduce

operator involvement by automating certain aspects

of low-level operations such as threshold selection,

histogram analysis, classification, etc. Generally,

these methods only use local information for each

pixel and do not incorporate information about the

shape of objects and their boundaries. Edge-based

methods rely on the evolution of a curve based on

internal and external forces, such as image gradient,

to delineate the boundary of the object structure

under analysis or pathology.

Pixel thresholding methods are more frequently

employed for image binarization, [13], [14], [15].

These methods are straightforward, and require

minimal computational costs, but are effective only

when all objects and backgrounds in the images are

clearly distinguished in terms of color or grayscale.

Hence, it is necessary to control the segmentation

process and adjust the results interactively.

In this context, in the article [16], a

segmentation of medical images is proposed based

on morphological operators in combination with

threshold value selection. In [17], morphological

operations were also used in combination with

threshold and division-based segmentation.

Furthermore, fuzzy methods are often employed for

image segmentation, such as Fuzzy C-Means

clustering (FCM), [18], [19].

In [17], for brain tumor extraction, an approach

based on morphological operations with threshold-

based segmentation and watershed line (LPE)

methods was utilized. Additionally, the Fuzzy C-

Means method, [18], [19], is widely used in medical

image segmentation. This method aims to separate

objects from each other and the background in the

image by extracting contours or segmenting them

into homogeneous regions.

In [20], to detect tumor boundaries in MRI

images for various cases of brain tumors, a hybrid

approach was proposed, combining the watershed

method and the Canny edge detection method. One

of these methods for representing medical image

contours uses color encoding, [21].

3 Thresholding Methods

This pertains to a fundamental method in image

segmentation, [22]. The general principle of

thresholding involves finding an appropriate

threshold value and then classifying all pixels in the

image based on the value of their grayscale levels

compared to this threshold to separate the regions of

interest from the image background. Some threshold

determination methods are based on parameters

other than grayscale level, such as entropy, [23] or

Tsallis entropy, [24]. For example, [25], formulated

the image thresholding problem as an iterative

discriminant analysis problem that allows for the

selection of an optimal threshold value. The

criterion used for threshold selection is based on

maximizing a statistical measure of separation

between classes. In all cases, the threshold obtained

through the methods mentioned above is ultimately

used for pixel classification based on their grayscale

levels.

In general, thresholding methods can be classified

into two categories:

• Global thresholding methods: These

methods are widely used in the segmentation of

mammographic images, [26], to detect tumor areas

or microcalcifications. The principle of these

methods is to determine the threshold value using

the overall information contained in the image. This

information is often presented in the form of a

histogram of grayscale levels in the image, [27],

[28]. Despite their widespread use, global

thresholding is not very effective in accurately

identifying regions of interest. In reality,

mammographic images represent the projection of a

3D scene onto a 2D observation space. This

projection results in significant overlaps of regions

that make up the breast tissue, [29], which limits the

effectiveness of these methods.

• Local thresholding methods: These

methods aim to locally refine the threshold value to

better identify regions of interest. The threshold

value is determined by considering only the

information contained in the local neighborhood of

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2023.19.23

El Fahssi Khalid, Ounasser Saida,

Mohamed Taj Bennani, Abenaou Abdenbi

E-ISSN: 2224-3488

216

Volume 19, 2023

each pixel, [29]. These methods have often

demonstrated better detection efficiency compared

to global thresholding methods. It's worth noting

that local thresholding methods have not only been

used for image segmentation but have also been

utilized as a preprocessing step for other algorithms,

such as those based on Markov fields, [30].

4 The Proposed Method

In [31], the approach used has led to improved

segmentation compared to classical methods.

However, the complexity of the method lies in the

selection of the initial contour point.

The major disadvantage of the level set method

is that it requires considerable thought to construct

appropriate speeds to progress the level set function.

Very high computational complexity and often very

slow convergence.

Among the disadvantages of the approach is the

lack of a reset term. In reality, the reset is a very

important step for the evolution of the contour, on

the one hand. On the other hand, the approach uses

an algorithm with very high computational

complexity and often very slow convergence, which

limits its use in real-time applications. Indeed, this

algorithm requires updating and calculating the

function for all points in the image and not just for

the zero level curve.

In this article, we propose a segmentation

approach based on the enhancement of the Otsu

thresholding and fuzzy C-means clustering, as

illustrated in Figure 1. The proposed method has

been applied for breast tumor detection.

For contour detection, various algorithms can be

used, such as Roberts, Prewitt, and Sobel, and more

complex ones like wavelet and Canny. In this

article, we used the Canny operator for contour

detection as it is considered to be a more efficient

tool for contour extraction.

After segmenting the image into a series of

homogeneous classes using the proposed

segmentation method, we apply the Canny operator.

The steps of this method can be described as

follows. In the first step, a Gaussian filter is used to

smooth the original images. Additionally, for each

pixel in the image, we calculate the magnitude and

direction of the gradient. The next step involves

selecting edge pixels (extreme pixels). A pixel is

considered an edge pixel if its gradient value is

greater than that of its two neighbors in the direction

of the gradient.

Fig. 1: Functional Diagram of the Proposed Method

However, the traditional Canny operator only

extracts gradients in the x and y directions within

2x2 neighborhoods. This method loses some

important contour information in the slope direction

(45° and 135°). To address this issue, we use a

method that calculates information about the

gradient magnitude and direction to determine the

gradient module of the pixel. This method takes into

account both localization accuracy and noise

immunity. For each pixel (i, j) in an image I, the

partial derivatives with respect to x and y for 45°

and 135° are as follows:

x (,) = ( + 1,) − ( − 1,),

y (,) = (, + 1) − (, − 1),

45° (,) = ( − 1, + 1) − ( − 1, − 1),

135° (,) = ( − 1, + 1) − ( + 1, − 1).

Enhanced fuzzy clustering

S= (S1+S2) / 2

Contour determination using

threshold (S)

Image analysis and

segmentation

Non-infected zone

Image reading

Pre-processing:

Filtering

Contrast enhancement

Fuzzy clustering (S2)

Otsu threshold (S1)

Infected zone

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2023.19.23

El Fahssi Khalid, Ounasser Saida,

Mohamed Taj Bennani, Abenaou Abdenbi

E-ISSN: 2224-3488

217

Volume 19, 2023

The gradient magnitude is:

󰇛󰇜



󰇛󰇜

󰇛󰇜

󰇛󰇜

󰇛󰇜

(1)

Gradient direction:

󰇛󰇜 󰇛󰇜

󰇛󰇜 (2)

5 Informative Features Extraction

The certainty of object classification in the

segmented image is strongly linked to the method

used for extracting informative features. In this

article, we proposed a classification algorithm based

on the application of Discrete Wavelet Transform

(DWT) and the Gray-Level Co-occurrence Matrix

(GLCM). The coefficients obtained through discrete

wavelet transform are analyzed to extract some of

the most informative statistical features, such as

energy, contrast, correlation, entropy, homogeneity,

and others.

Finally, the procedure for diagnosing

mammography images involves using a machine

learning algorithm for the classification of tumor-

infected areas (benign or malignant). The

segmentation method proposed in this article has

yielded good results, making it easier to choose a

simple and effective learning algorithm, such as the

Support Vector Machine (SVM) method.

6 Results and Discussion

In this experimental study, we used the database

(mini-MIAS), [32], for mammography analysis

(breast cancer).In this work, we formed two classes

of data with 100 images each of size 256 x 256. One

class consists of healthy breast MRI images. While

the second consists of MRI images with

abnormalities. The two classes are used for learning

the diagnostic system. An additional breast cancer

dataset consisting of 50 images is used to test the

system. The results of the experimental study of

breast tumor segmentation and detection are shown

in Figure 2.

We can also notice that the brightness and

contrast of the image vary from one image to

another (The upper part of the figure from left to

right). The first column (from left) represents the

original images, and the second column represents

the image results after the preprocessing step.The

third and last column illustrates the result after

masking and processing (final edge images).

Comparing the proposed method with the methods

[28], [29] for identifying and segmenting a breast

tumor presented, one can notice the developed

algorithms capable of segmenting tumors and

normal regions better than other methods.Indeed ,

the results obtained in [28], [29] include a non-

tumor region in the segmented image, which

represents an ambiguity for the diagnosis (without

outline of the tumors). The proposed method

successfully detects the tumor region (as well as the

normal breast region in the original image) as shown

in Figure 2.

Fig. 2: Experimental result of the proposed method for breast tumor detection

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2023.19.23

El Fahssi Khalid, Ounasser Saida,

Mohamed Taj Bennani, Abenaou Abdenbi

E-ISSN: 2224-3488

218

Volume 19, 2023

7 Conclusion

The experimental results obtained show that the

effectiveness of the proposed method of

segmentation by groups. The effectiveness of the

approach lies in the combination of two

approaches, the Otsu thresholding method and

fuzzy clustering.The application of the discrete

wavelet transform (DWT) and the GLCM co-

occurrence matrix ensured an improvement in the

analysis quality during the extraction of image

features.

In comparison with other segmentation

methods, the new threshold used in the proposed

approach significantly increased the precision and

quality of medical image segmentation. Which will

subsequently contribute to improving the

diagnostic results of MRI images while detecting

breast tumors.

References:

[1] Lo SCB, Chan H-P, Lin JS, Li H, Freedman

M, Mun SK. Artificial convolution neural

network for medical image pattern

recognition, Neural networks, 1995. Т. 8., №.

7-8, С. 1201-1214.

[2] Guotai Wang, Wenqi Li, Sebastien

Ourselin, Tom Vercauteren. Automatic brain

tumor segmentation using cascaded

anisotropic convolutional neural networks,

International MICCAI brain lesion

workshop, Springer, Cham, 2017, С. 178-

190.

[3] Xiaojian Kang, Timothy J Herron, And U

Turken, David L Woods. Diffusion

properties of cortical and pericortical tissue:

regional variations, reliability, and

methodological issues, Magnetic Resonance

Imaging, 2012. – Т. 30. – №. 8. – С. 1111-

1122.

[4] Mariana Leite, David Gobbi, Marina

Salluzi, Richard Frayne, Roberto

Lotufo, Letícia Rittner. 3D texture-based

classification applied on brain white matter

lesions on MR images, Medical Imaging

2016: Computer-Aided Diagnosis. –

International Society for Optics and

Photonics, 2016. – Т. 9785. – С. 97852N.

[5] Schultz T., Theisel H., Seidel H. P.

Topological visualization of brain diffusion

MRI data, IEEE Transactions on

Visualization and Computer Graphics, 2007.

Т. 13., №. 6. – С. 1496-1503.

[6] J. Singh, H Kaur, Plant Disease Detection

Based on Region-Based Segmentation and

KNN Classifier, International Conference on

ISMAC in Computational Vision and Bio-

Engineering 2018 (ISMAC-CVB).

[7] Bahadir Karasulu, Serdar Korukoglu, A

simulated annealing-based optimal threshold

determining method in edge-based

segmentation of grayscale images, Applied

Soft Computing Journal, Vol. 11, Issue 2, p.

2246-2259, 2011

[8] Natthan Singh, Shivani Goyal, Determination

and Segmentation of Brain Tumor Using

Threshold Segmentation with Morphological

Operations, Soft Computing: Theories and

Applications, 2018, Vol. 584.

[9] Weibei, Dou, Su, Ruan, Yanping, Chen,

Daniel, Bloyet, and Jean-Marc, Constans.

(2007) “A framework of fuzzy information

fusion for segmentation of brain tumor

tissues on MR images.” Image and Vision

Computing 25: 164-171.

[10] Gul, Moonis, Jianguo, Liu, Jayaram K.,

Udupa, and David B., Hackney. (2002)

“Estimation of tumor volume with fuzzy-

connectedness segmentation of MR images”.

American Journal of Neuroradiology, 23:

352-363.

[11] Meritxell, Bach Cuadra, Claudio, Pollo,

Anton, Bardera, Olivier, Cuisenaire Jean-

Guy Villemure, and Jean-Philippe, Thiran

(2004) “Atlas-based segmentation of

pathological MR brain images using a model

of lesion growth.” IEEE Transactions on

Medical Imaging 23 (10): 1301-1313.

[12] Jason, J. Corso, Eitan, Sharon, and Alan,

Yuille. (2006) “Multilevel segmentation and

integrated Bayesian model classification with

an application to brain tumor segmentation”,

in MICCAI2006, Copenhagen, Denmark,

Lecture Notes in Computer Science, 4191,

Springer, Berlin: 790-798.

[13] Fatma Taher, Naoufel Werghi and Hussain

Al-Ahmad, “A thresholding approach for

detection of sputum cell for lung cancer early

diagnosis” Vol. 2012, Issue 600 CP2012 IET

Conference on Image Processing, July 2012.

[14] Gul, Moonis, Jianguo, Liu, Jayaram K.,

Udupa, and David B., Hackney. (2002)

“Simple and efficient method of low-contrast

grayscale image binarization”. Vol. 9972

LNCS, p 142 – 150, International Conference

on Computer Vision and Graphics, ICCVG

Warsaw, September 2016.

[15] An Algorithm for fast adaptive image

binarization with applications in radiotherapy

imaging, IEEE Transactions on Medical

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2023.19.23

El Fahssi Khalid, Ounasser Saida,

Mohamed Taj Bennani, Abenaou Abdenbi

E-ISSN: 2224-3488

219

Volume 19, 2023

Imaging, Vol. 22, Issue 1, pp.22-28, January

2003.

[16] Priya and Vivek Singh Verma, New

Morphological Technique for Medical Image

Segmentation, 3rd IEEE International

Conference on "Computational Intelligence

and Communication Technology" (IEEE-

CICT 2017).

[17] DiyaChudasama, Tanvi Patel and Shubham

Joshi "Image Segmentation using

Morphological Operations", International

Journal of Computer Applications (0975 –

8887), Vol. 117 – No. 18, May 2015.

[18] Chuang, K.-S., Tzeng, H.-L., Chen, S., Wu,

J., & Chen, T.-J. Fuzzy c-means clustering

with spatial information for image

segmentation. Computerized Medical

Imaging and Graphics, 30(1), 9–15. 2006.

[19] Huiyu Zhou, Gerald Schaefer and Chunmei

Shi, Fuzzy C-Means Techniques for Medical

Image Segmentation. Fuzzy Systems in

Bioinformatics and Computational Biology,

pp. 257–271, 2009.

[20] Norhafiza Hamzah, Mohd Sapaweh Asli and

Rechard Lee, Skin Cancer Image Detection

using Watershed Marker-Controlled and

Canny Edge Detection Techniques.

Transactions on Science and Technology,

Vol. 5, No. 1, 1 - 4, 2018.

[21] Xianhua Zeng, Shiyue Tong, Yuzhe Lu,

Liming Xu and Zhiwei Huang, Adaptive

Medical Image Deep Color Perception

Algorithm, IEEE Access, Vol. 8, 2020.

[22] R. C. Gonzalez and R. E. Woods: Digital

Image Processing. 2ed, Prentice Hall, 2002.

[23] Lo S. C. B. et al. Artificial convolution

neural network for medical image pattern

recognition, Neural networks, 1995 Т. 8 №.

7-8 С. 1201-1214.

[24] J. Mohanalin, P. K. Kalra and N. Kumar:

Tsallis Entropy Based Contrast Enhancement

of Micro calcifications, International

Conference on Signal Acquisition and

Processing, ICSAP, pp.3-7, 2009.

[25] M. A. Duarte, A. V. Alvarenga, C. M.

Azevedo, A. F. Infantosi and W. C. Pereira:

Automatic microcalcifications segmentation

procedure based on Otsu's method and

morphological filters, Health Care

Exchanges (PAHCE), pp. 102-106, 2011.

[26] H. Cheng, X. Shi, R. Min, L. Hu, X. Cai, and

H. Du: Approaches for automated detection

and classification of masses in

mammograms, Elsevier, vol. 39, no.4, pp.

646-668, 2006.

[27] G. M. Victor, M.G. Daniel, and R. Juan:

Iterative method for automatic detection of

masses in digital mammograms for

computer-aided diagnosis, Proc. SPIE, vol.

3661, pp. 1086-1093, 1999.

[28] T. Matsubara, T. Fujita, H. Kasai, S. Goto,

M. Tani, Y. Hara, and T.T. Endo:

Development of new schemes for detection

and analysis of mammographic masses,

IEEE. Comput. Soc, pp. 63-66, 1997.

[29] L. Li, W. Qian, L. P. Clarke, R. A. Clark, and

J. A. Thomas: Improving mass detection by

adaptive and multiscale processing in

digitized mammograms, Proc. SPIE, vol.

3661, pp. 490-498, 1999.

[30] H. D. Li, M. Kallergi, L. P. Clarke, V. K.

Jain, and R. A. Clark: Markov random field

for tumor detection in digital mammography,

IEEE. Transactions on Medical Imaging, vol.

14, no. 3, pp. 565-576, 1995.

[31] Liu, J., Wei, X. & Li, L. MR image

segmentation based on level set method.

Multimed. Tools Appl. 79, 11487–11502,

2020.

[32] The mini-MIAS database of mammograms.

http://peipa.essex.ac.uk/info/mias.html

(Access Date: September 7, 2023).

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The authors equally contributed in the present

research, at all stages from the formulation of the

problem to the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflicts of interest to declare.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.e

n_US

WSEAS TRANSACTIONS on SIGNAL PROCESSING

DOI: 10.37394/232014.2023.19.23

El Fahssi Khalid, Ounasser Saida,

Mohamed Taj Bennani, Abenaou Abdenbi

E-ISSN: 2224-3488

220

Volume 19, 2023