Segmentation Method of Skin
MRI
High Resolution in vivo
Type of article: Original article.
Rachida Zegour1,Ahror
Belaid2, Douraied Ben Salem3,4
1.Ph.D. student at Medical Computing Laboratory (LIMED), Faculty of Exact
Sciences,University of Abderrahmane Mira of Bejaia, Algeria.
2. Assistant professorat Medical Computing Laboratory (LIMED), Faculty of
Exact Sciences,University of Abderrahmane Mira of Bejaia, Algeria.
3. Professor at INSERM UMR 1101 Laboratory of Medical Information
Processing (LaTIM), 5 avenue Foch, 29200 Brest, France.
4 .Professor at Neuroradiology Department, CHRU-Brest, boulevard
Tanguy-Prigent,
29609 Brest, France.
Abstract:
Background: Recently, Magnetic
Resonance Imaging (MRI) has been used in clinical
application as non-invasive medical modality; it is rarely used to study the
anatomy physiological, and biochemical of the skin, in spite of its very
attractive modality for skin imaging. It makes an ideal imaging modality of
unique soft tissue contrast to study the skin anatomy and to observe the different
skin’s layers. However, MRI provides
a big data with high quality. The analysis of these data requires computerized
methods to help clinicians and to improve diagnosis. Several image
processing methods have been used by doctors to facilitate qualitative
diagnosis, segmentation is one of these methods used in clinical applications
in order to understand medical data and extract useful information. This study
aims to use the segmentation method in order to observe the anatomy of skin
layers before and after applying moisturizer.
Methods: We will classify
segmentation approaches for MRI data into three basics classes: Edge based
segmentation; Region based segmentation, and Thresholding segmentation. Then we
will briefly describe Fuzzy C-means Clustering method, which is used to segment
our MR data represented by a sample of thirty-five (35) healthy.
Results: We have observed the
skin’s layers before and after applying moisturizer topic of the feet as a
result of the FCM segmentation method.
Our study showed that FCM is
an efficient Algorithm used for medical images, because of its fuzzy nature;
also it gives images segmentation results with different classes of skin layers
before and after the application of moisturizer topic.
Conclusion: MRI is an
attractive modality to study the human skin; it makes an ideal observation of
the different skin layers in vivo. However, the segmentation of MRI data by FCM
clustering is a computerized method to help clinicians in order to study the
skin anatomy.
Keywords: MRI High
Resolution, segmentation, FCM.
Corresponding author: Rachida Zegour,
Medical Computing Laboratory (LIMED), Faculty of Exact Sciences, University of
Abderrahmane Mira of Bejaia, Algeria Email: zegourr@gmail.com
Received: Mai 30,
2018, Accepted: September 02, 2018, English editing: September 28, 2018, Published:
September 30, 2018.
Screened by
iThenticate.©2017-2018 KNOWLEDGE KINGDOM PUBLISHING.
1. Introduction
Over the last years,
Medicine has used many image modalities in order to make diagnostic easier.
Magnetic Resonance Imaging (MRI) is one of non-invasive modalities which uses
the distribution of water in human body in order to produce medical images.
MRI modality provides a big
data with high quality, the analysis of these data to extract important
information manually becomes a complex task for clinicians, it is time
consuming and it may cause errors because of the large and various studies.
However, MR data analysis requires a set of computerized methods to improve diagnosis
and make it easy [2].
Nowadays, many computerized
methods assist doctors for MR data diagnostic. Segmentation method is most used
in MR data processing for many applications in order to understand medical data
and extract useful information.
Hence, MRI modality is
rarely used in clinical application to study the anatomy physiological, and
biochemical of the skin [1], but is a very attractive modality for skin
imaging. It makes an ideal imaging modality of unique soft tissue contrast to
study the skin anatomy and to observe the different skin’s layers [6].
Skin covers the total area
of the human body, it is composed of three layers [3]: (1) Epidermis: the
outermost layer of skin provides a waterproof barrier, which is composed of
approximately 95% of keratinocytes, with four layers: basal, spinous, granular
and corneum layers. These layers are undergoing continuous transformation and
form the remaining 5% [4]. (2) Dermis: is beneath the epidermis, which is
composed of fibroblasts, dermal dendritic cells, mastocytes and macrophages
(70%). Other components include collagen, elastic fibers and air follicles. The
dermis is grouped into two categories: papillary dermis and reticular dermis
[4]. (3) Fat: The deeper subcutaneous tissue is made of fat and connective
tissue [3].
Water distribution in human skin
plays a key role indifferent skin functions, such as thermoregulation, barrier
function…etc. [5]. Magnetic Resonance Imaging (MRI) uses magnetic proprieties
of some atoms like Hydrogen (1H), Sodium (23Na), Carbon (13C)..etc.
As the human body is made of 70% water (H2O), MRI in clinical
applications, uses magnetic properties of Hydrogen atoms to produce medical
images [7].
This paper will address a
medical challenge of in vivo high-resolution skin MR imaging, our data is skin
MR images has been investigated using 3-Tesla scan. With high resolution to
make the promising of skin in vivo approaches possible, and also allows
observing the anatomical and the physiological properties of the skin with fine
exploration [1]. Our data have been processed using the segmentation method in
order to observe the skin anatomy of the feet before applying moisturizer topic
and one hour after that. We will also review in this paper the methods of MR
images segmentation, then; we will present their classification with a detailed
description of Fuzzy C-means Clustering (FCM) method. Furthermore, we will give
some related works used FCM algorithm with MRI images. Then we will describe
our study's sample, and come up to discuss the results. Finally, we will
conclude the review by a conclusion.
2. Methods
Segmentation is one of the
most used methods to process images with several approaches. In this paper, we
present a categorization of three classes for medical image methods (see figure
1) [8]: (a) Edge-based segmentation: consists of finding boundaries between
regions based on detecting sharp and local changes in the intensity value, with
three interested characteristics: edge, line and isolated point. (b)
Region-based segmentation: based on a set of intensity attributes value to find
the similar region directly. This approach involves many methods such as region
growing, split, merge and clustering method. (c) Thresholding method: is one of
the most applied approaches in image processing, because of its properties,
simplicity, and its speed of computation. It classifies each pixel of image
according to the specific threshold value.
Other approved methods have
been developed such as morphological watersheds, active contours and level set
method. The following diagram represents the classification methods of
segmentation:
Figure1: Classification of segmentation methods
Fuzzy C-means Clustering
(FCM) is a clustering method used to segment images, it assigns similar pixels
into the same class using fuzzy membership.
Let X=(x1 , x2,
...... ,xn ) represents an image with n pixels, which is splitted into
C clusters. Where, each xi represents a set of pixel features. FCM
algorithm minimizes the following objective function [11]:
/
Where, uij is the
membership of pixel xj in the ith cluster, vi represents
the ith cluster center, ||.|| is a norm metric, and m is a parameter
which defines the fuzziness of the resulting clusters.
FCM algorithm assigns a high
membership to the pixels when they are close to the centroid of the cluster and
a low membership if not, in order to minimize the objective function J. The
probability value of pixels depends solely on the distance between the pixel
and each individual cluster center. FCM algorithm starts with an initial guess
for each cluster center, then update membership and cluster centers for each
iteration as follows [11]:
When the changes in the
membership function or the cluster center at two successive iteration steps are
constant, FCM algorithm converges to a solution with vi representing
a saddle point of the objective function [11].
FCM is used for medical
images because of its fuzzy nature, where one pixel can belong to more than one
cluster according to different probability value, and with a higher performance
than crisp methods [9]. For example, some developers [11- 13] used spatial FCM
algorithm for MRI data to improve the robustness of the conventional algorithm.
Others [14] used FCM algorithm to segment MRI images.
Fuzzy C-means Clustering
(FCM) method has two problems: The first one is its less performance with
imaging noise, and the second is the Euclidean distance which is used [10].
Because of these problems, many variations to the FCM have been developed like
Bias-Corrected Fuzzy C-Means (BCFCM), Possibilistic Fuzzy C-Means (PFCM) ...
etc [9].
3. Results and discussion
Our data are the results of
a study conducted between September 2014 and August 2015 in the research center
at the university Hospital in Brest, France [1]. The study performed an MRI of
the feet before and after applying moisturizer topic. The study's sample was
thirty-five healthy volunteers comprising 17 women and 18 men with an average
age of 25.3 years
For each subject, we had MRI sequences of images
with 8echoes at 3T. The following images are example:
Figure2: MRI sequence: Turbo Spin Echo (TSE)
calculation T2-weighted sequence.
The T2 calculation of TSE sequence using Fiji ImageJ gives these images
as results:
Figure 3: T2 Calculation Results (a) Before applying moisturizer topic
(b) After application.
The application
of Fuzzy C-means Clustering using MATLAB for 10 clusters gives these results:
Figure 4: Results of FCM
segmentation method; (a) Before applying moisturizer topic (b) After
application.
This study involves applying
FCM segmentation method for MRI data, to observe the effects of a topical
moisturizer on the different skin layers. We can observe that the epidermis
layer before applying the moisturizer topic is represented by different classes
than after application.
We can conclude first that
MRI makes an ideal imaging modality of unique soft tissue contrast to study the
skin anatomy and to observe the different skin’s layers.
Then, the application of
moisturizer topic provides effects in epidermis layer of skin, exactly in
stratum corneum, which view with large area after the application of
moisturizer topic.
4. Conclusion
The aim of this
article is to review the segmentation approaches for Magnetic Resonance
Imaging. We started by classifying these approaches, then briefly described
Fuzzy C- means Clustering method. Furthermore, we presented the sample of our
study with a discussion of the results obtained using FCM segmentation method.
We conclude that MRI is an attractive modality to study the skin water content,
it makes an ideal observation of the different skin layers in vivo with three
dimensions. However, the segmentation of MRI data by FCM clustering is a
computerized method to help clinicians in order to measure skin hydration.
5. Conflict of interest statement
Authors declare no conflicts of interest.
6. Authors’ biography
Rachida ZEGOUR is currently a Ph.D. student in Medical Computing Laboratory (LIMED)
at Mira Abderahmen university, Bejaia, Algeria. She obtained her Master’s
degree in 2015 from M’Hamed Bouguerra University, Boumerdes, Algeria.
Currently, she is working on segmentation of medical images (MRI) using Fuzzy
methods. Her main research interests include medical image processing and machine
learning.
Dr. Ahror BELAID received the M. S. degree in Operation Research from University of
Abderrahmane Mira, Bejaia, Algeria. He received his Ph.D. degree at the
Department of Information Processing Engineering, Heudiasyc laboratory,
Compiègne University of Technology, France. He is currently Associate Professor
with the Department of Computer Science, Abderrahmane Mira University. His
current research interests include Image Processing and Machine Learning.
Douraied BEN SALEM, MD, PhD, is Professor of Radiology at the University of Western
Brittany (UBO, Brest, France) and member of the Laboratory of medical
information processing —LaTIM, INSERM UMR 1101 (Brest, France). He received his
medical and doctoral degree from the University of Burgundy (Dijon, France). He
is Associate Editor of the "Journal of Neuroradiology" and member of
the Editorial Board of "Heliyon" and of the "Journal of Forensic Radiology and
Imaging".
7. References:
[1] Jawad Mesrar, Julien Ognard, Marc Garetier,
DavidChechin, Laurent Misery and Douraied Ben Salem, In vivo skin moisturizing
measurement by high resolution 3 Tesla magnetic resonance imaging, Skin
Research and Technology 2017; 23: 289–294, doi: 10.1111/srt.12333.
https://doi.org/10.1111/srt.12333
[2]Ivana Despotovic, Bart Goossens, and Wilfried
Philips, MRI Segmentation of the Human Brain;Challenges, Methods, and
Applications, Computational and Mathematical Methods in Medicine, Volume 2015
(2015), Article ID 450341, 23 pages http://dx.doi.org/10.1155/2015/450341.
https://doi.org/10.1155/2015/450341
[3]J.A. McGrath, R.A.J. Eady & F.M. Pope, Anatomy
and Organization of Human Skin, textbook of dermatology 1-1-153, 2010,
https://doi.org/10.1002/9780470750520.ch3.
https://doi.org/10.1002/9780470750520.ch3
[4]Elisa de Oliveira Barcaui, Antonio Carlos Pires
Carvalho, Juan Pi-eiro-Maceira, Carlos BaptistaBarcaui, and Heleno Moraes,
Study of the skin anatomy with high-frequency (22 MHz)ultrasonography and
histological correlation, Radiol Bras. 2015 Sep-Oct; 48(5): 324–329.
doi:10.1590/0100-3984.2014.0028.
https://doi.org/10.1590/0100-3984.2014.0028
[5]Lídia Palma, Liliana Tavares Marques, Julia Bujan,
and Luís Monteiro Rodrigues, Dietary water affects human skin hydration and
biomechanics, Clin Cosmet Investig Dermatol. 2015; 8: 413–421. Published online
2015 Aug 3. doi: 10.2147/CCID.S86822.
https://doi.org/10.2147/CCID.S86822
[6]Joëlle K. Barral, Neal K. Bangerter, Bob S. Hu, and
Dwight G. Nishimura1, In Vivo High-Resolution Magnetic Resonance Skin Imaging
at 1.5 T and 3 T, published in final edited form as: Magn Reson Med. 2010 Mar;
63(3): 790–796. doi: 10.1002/mrm.22271.
https://doi.org/10.1002/mrm.22271
[7]Benoit Scherrer, Segmentation of tissues and
structures on brain MRI s: agents local Markovians cooperatives and Bayesian
formulation, 29 October 2008
[8] Rafael C. Gonzalez and Richard E. Woods, Digital
image processing, Forth edition, 2017
[9]Mahipal Singh Choudhry and Rajiv Kapoor,
Performance Analysis of Fuzzy C-Means Clustering Methods for MRI Image
Segmentation, Twelfth International Multi-Conference on Information
Processing-2016 (IMCIP-2016).
[10] Yuhui Zhenga, Byeungwoo Jeond, Danhua Xua, Q.M.
Jonathan Wua, and Hui Zhanga, Image segmentation by generalized hierarchical
fuzzy C-means algorithm, Journal of Intelligent & Fuzzy Systems 28 (2015)
961–973 DOI:10.3233/IFS-141378 .
[11] Keh-Shih Chuang, Hong-Long Tzeng, Sharon Chen,
Jay Wu and Tzong-Jer Chen, Fuzzy c-means clustering with spatial information
for image segmentation, Computerized Medical Imaging and Graphics 30 (2006)
9–15.
https://doi.org/10.1016/j.compmedimag.2005.10.001
PMid:16361080
[12] Mohamed N. Ahmed, Member, IEEE, Sameh M. Yamany,
Member, IEEE, Nevin Mohamed, Aly A. Farag, Senior Member, IEEE, and Thomas
Moriarty, A Modified Fuzzy C-Means Algorithm for Bias Field Estimation and
Segmentation of MRI Data, IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 21, NO. 3,
MARCH 2002 .
[13] Sudip Kumar Adhikari , Jamuna Kanta Sing, Dipak
Kumar Basub, Mita Nasipuri, Conditional spatial fuzzy C-means clustering
algorithm for segmentation of MRI images, Applied Soft Computing 34 (2015)
758–769.
https://doi.org/10.1016/j.asoc.2015.05.038
[14] Mohamed Baghdadi,
NacéraBenamrane and Lakhdar Sais, Fuzzy generalized fast marching method for 3D
segmentation of brain structures, Imaging Systems and Technology, Volume27,
Issue3, September 2017, Pages 281-306.
https://doi.org/10.1002/ima.22233