Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

RESEARCH

Open Access

Spatial based Expectation Maximizing (EM)

M A Balafar

Abstract
Background: Expectation maximizing (EM) is one of the common approaches for image segmentation.
Methods: an improvement of the EM algorithm is proposed and its effectiveness for MRI brain image
segmentation is investigated. In order to improve EM performance, the proposed algorithms incorporates
neighbourhood information into the clustering process. At first, average image is obtained as neighbourhood
information and then it is incorporated in clustering process. Also, as an option, user-interaction is used to improve
segmentation results. Simulated and real MR volumes are used to compare the efficiency of the proposed
improvement with the existing neighbourhood based extension for EM and FCM.
Results: the findings show that the proposed algorithm produces higher similarity index.
Conclusions: experiments demonstrate the effectiveness of the proposed algorithm in compare to other existing
algorithms on various noise levels.
Keywords: Em, Segmentation, Neighbourhood

1. Background
The application of image processing techniques for medical
imaging process rapidly increases. Most medical images are
stored and represented in softcopy [1]. Ultrasound, X-ray
computed tomography, digital mammography and magnetic resonance imaging (MRI) are the most common
medical imaging types [2]. MRI can give different grey level
for different tissues and various types of neuropathology if
its acquisition parameters are adjusted [3].
Data acquisition, processing and visualization techniques facilitate diagnosis. Medical image segmentation
plays a very important role in many computer-aided
diagnostic tools. These tools could save clinicians time
by simplifying complex time-consuming processes [4].
The main part of these tools is to design an efficient segmentation algorithm. Medical images mostly contain
unknown noise [5], in-homogeneity [6] and complicated
structures. Therefore, segmentation of medical images is
a challenging and complex task. Medical image segmentation has been an active research area for a long time.
There are many segmentation algorithms but there is no
generic algorithm for a totally successful segmentation of
medical images [7].
Correspondence: [email protected]
Dept of IT, Faculty of Electric and Computer, University of Tabriz, Tabriz, East
Azerbaijan, Iran

Clustering methods are common for MRI brain segmentation. Expectation-maximization (EM) and fuzzy c-mean
(FCM) are the most popular clustering algorithms. The
Gaussian mixture model (GMM) is a popular segmentation method. EM is used to estimate the parameters of
this model. FCM and EM only consider the intensity of
images and in noisy images, intensity is not trustful [8-10].
Usually, spatially adjacent pixels belong to the same cluster. Many algorithms introduced to make FCM [11-17]
and EM robust against noise but nevertheless most of
them were and are flawless to some extent. Usually, spatially adjacent pixels belong to the same cluster. Many
researchers attempted to incorporate spatial information
into FCM and EM to overcome the noise problem. Zhang
et. al. [18] proposed a novel Gaussian hidden Markov
Random Field (HMRF) model to integrate spatial information into Gaussian model. They used a Markov Random
Field-Maximum A Posteriori (MRF-MAP) approach to
estimate the model solution. Recently, Tang et al. [19]
proposed a neighbourhood-weighted Gaussian mixture
model to overcome misclassification on the boundaries
and on inhomogeneous regions of MRI brain images with
noise. A. R. F. d. Silva [20] proposed two Bayesian algorithms (DPM, rjMCMC) which use Markov chain sampling techniques to find normal mixture models with an
unknown number of components. They used algorithms
for MRI segmentation and compared performance of their

2011 Balafar; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

algorithms with published results for two exist Bayesian

based MRI brain segmentation methods (KVL [21], MPMMAP [22]).
Gonzlez Ballester et al. [23] and Tohka et al. [24]
reported a statistical models namely a novel trimmed
minimum covariance determinant (TMCD) for the estimation of the parameters of partial volume models to
address partial volume averaging.
In order to make Gaussian mixture model more robust
against complex tissue spatial layout, Greenspan et al. [25]
proposed the parameter-tied, constrained Gaussian mixture model (CGMM) to capture this problem. The mixture model composed of a large number of Gaussians for
each tissue is used to capture the complex tissue spatial
layout. The Gaussian parameters of a tissue are tied using
intensity as global feature. The parameters are learned
using the expectation-maximization (EM) algorithm.
In [26], a nonparametric Bayesian model, known as
Dirichlet process mixture model (DPMM) is proposed to
overcome the limitations of current parametric finite mixture models. The DPMM permits unknown number of
components in the mixture and allow robust segmentation
of brain with unknown or incomplete specifications.
In [27], local cooperative unified segmentation (LOCUS)
approach based on distributed local MRF models for brain
segmentation is presented. The volume is partitioned into
sub volumes and a set of local and cooperative Markov
random field (MRF) models are distributed. In order to
ensure consistency, neighbour local MRFs are estimated
cooperatively. The intensity in-homogeneity correction is
not required due to precisely fit of Local estimation with
the local intensity distribution.
In this paper, a new modification to GMM and EM is
introduced by incorporating neighbourhood information
into likelihood function and EM steps. The average of
neighbour pixels around each pixel is calculated prior to
GMM clustering and incorporated in GMM and EM functions beside the pixel value.
The rest of this paper is organized as follows. The standard GMM model and EM segmentation algorithm are
presented in Section 2.1. In Section 2.2, proposed modified
EM algorithm is described. Also, improvement of segmentation results using use-interaction is presented in section
2.3. Experimental and comparison results are presented in
Section 3 and this paper is concluded in Section 4.

Page 2 of 14

(voxel) with M mixing coefficients. Each component is

assigned to one target class and the goal is to obtain the
class probabilities of each pixel (voxel). The probability
distribution of the jth component is denoted by pj(xi|j),
where xi is pixel i in input image and j is the parameter
(mean j and covariance matrix j) of component j. The
probability distribution each pixel (voxel) can be
described as a mixture of probability distributions as
follows:
p(xi | ) =

M


j pj (xi | j )

j=1

(1)

=
e
det(2 j )

Where a j denotes the mixture coefficient with the

M

j = 1 The probability distribution of
constraint,
j=1

component j is modelled by a Gaussian distribution

with mean j and covariance matrixj:
pj (xi | j ) = pj (xi | j , j )

2.1 Standard GMM

The Gaussian mixture model assumes M mixed component densities (Gaussian distribution) for each pixel

(2)

Usually, maximum likelihood (ML) estimation is used

to find the parameters. The log-likelihood expression for
the parameter and the image X is defined as follows:
log(L( | X)) = log

N


p(xi | )

i=1

N


log(

i=1

M


(3)

jt pj (xi |jt ))

j=1

Finding the ML solution from this equation is difficult.

Usually, the expectation-maximization (EM) is used to
obtain the parameters. EM steps are demonstrated in
the following:
E-step. Bayes rule is used to obtain the probability of
data xi belong to class j (E-step):
p(j | xi , ) =
t

jt pj (xi | jt )
M

j=1

2. Methods
A modification to GMM is introduced by incorporating
neighbourhood information into likelihood function and
EM steps.

(xj )T j1 (xj )/2

jt pj (xi | jt )

(4)

M-step. Probability obtained in E-step is used to

obtain mixing coefficient, mean and covariance matrix
(M-step):
jt+1 =

N
1 
p(j | xi , t )
N
i=1

(5)

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

N


t+1
j

Page 3 of 14

N


xi p(j | xi , t )

i=1
N


(6)
p(j | xi

t+1
j

((1 ) xi + x i )p(j | xi , t )

i=1
N


, t)

i=1

N


jt+1

i=1

p(j | xi , t ).(xi t+1

j )(xi
N


c. In Equation 7, the distance of x i from the component centre is added to the distance of xi from the component centre as neighbourhood information:

T
t+1
j )

(7)
p(j | xi

, t)

t+1 T
d(x) = (x t+1
j )(x j )
N


c. EM steps are repeated until convergence.

jt+1

p(j | xi , t ).(d(xi ) + .d(xi ))

i=1
N


2.2. Modified GMM

The average of neighbour pixels around x i is calculated

prior to GMM clustering. In the likelihood function
(Equation 3), distribution value of x i is added to the distribution value of pixel xi as neighbourhood information:
log(L( | X)) = log

N


p(xi | ) =

i=1

i=1

log(

M


p(j | xi

i=1

i=1

N


(10)
, t)

(8)

jt [(1 ) pj (xi | jt ) + pj (xi | jt )])

A = [(1 ) pj (xi | jt ) + pj (xi | jt )]

p(j | xi , t ) =

jt .A
 t
j .A
j=1

(9)

b. In Equation 6, x i is added to xi as neighbourhood

information:

p(j | xi , t )

i=1

In MRI, noise behaves as Rician distributed noise.

Rician noise approaches Gaussian distribution in high
Signal to Noise Ratio (SNR) and Rayleigh distribution in
low SNR [28]. Rician distribution in the background is
Rayleigh because there is no signal. The Rayleigh PDF
of the statistically independent observations is
n

Oi (O2 )/(2 2 )
p({Oi }) =
e i
2

(12)

i=1

j=1

The parameter b determines the weight of neighbourhood information. Incorporating neighbourhood information improves the performance of segmentation
methods in high level of noise, but the blurring effect
degrades the performance of them in low noise level. In
order to overcome the degrading effect of algorithms in
low level of noise, the variance of noise is used to specify the weight of neighbourhood information (b). Its
value is set to s, where s is the variance of noise. In
previous neighbourhood based EM extensions, neighbourhood information is calculated in clustering iteration; but in this algorithm x i is computed before
iteration, thus, the clustering will be faster. An extension
of EM named EM-1 is introduced to solve likelihood
function. The EM is modified as follows:
a. In Equation 4, distribution value of x i is added to the
distribution value of pixel xi as neighbourhood information:

(11)

2
is the variance of
Where O is observations and Noise
noise. The variance of noise is obtained by maximizing
the log-likelihood of PDF function with respect to variance:

1  2
Oi
2n
n

2
Noise
=

(13)

i=1

In other words, background pixels are considered as

observations (O) and the variance of noise is obtained
applying equation 13 on background pixels values. For
that, the powers of background pixels values are computed and half of the average of resulted values is considered as variance of the noise.
Also, in-homogeneity correction [6] is applied to input
image with in-homogeneity pollution and the propose
GMM is applied on in-homogeneity corrected image.
2.3. Improving Segmentation Results Using User
Interaction

Sometimes, due to in-homogeneity, low contrast, noise

and inequality of content with semantic, automatic methods fail to segment image correctly. Therefore, for these
images, it is necessary to use user interaction to correct
methods error [29]. However, robust semi-automatic
methods can be developed in which user interaction is
minimized.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 4 of 14

Moreover, reported results on simulated volumes for

existing extensions of EM (DPM, rjMCMC, KVL, MPMMAP) and existing neighbourhood based extension for
FCM (FCM_S [32], FCM_EN [33], FGFCM [34], FLICM
[35] and NonlocalFCM [36]) are used to evaluate proposed algorithm.
Also, the reported results on real volumes from IBSR
are used to evaluate proposed algorithms. Furthermore,
mentioned FCM extensions simulated and tested on real
volumes.
The results of algorithms are compared quantitatively
to analyse their performance. The neighbourhood size, N
for proposed algorithm is set to 3 3. Three indices
(similarity index, false positive ratio and false negative
ratio) [37] are used to evaluate the algorithms quantitatively. The similarity index ri of class i is the degree of
the class pixels matching between ground truth and segmentation result for the same class. The false positive
ratio rfp represents extra pixels of class i and the false
negative ratio rfn represents lost pixels of class i. They
are defined as follows:

Sometimes, segmented image, for example in Figure 1

(b), either has pixels from two or more tissues in one cluster (csf and grey matter of brain in cluster number 2) or
pixels from one tissue in two or more clusters(white matter in clusters number 2 and 3). For solving this problem,
user selects clusters contain several tissues (cluster number 1) to be re-clustered to two sub clusters. Figure 1(c)
demonstrates sub clusters of class number 1. The cluster
number 1 is clustered to two sub clusters number 11
and 12.
This process continues until user is satisfied. That
means quality of segmentation depends on user. Then,
to solve problem of several clusters for one tissue, user
selects clusters for each tissue (clusters 12 is also
selected for grey matter). Steps of this method listed as
follow:
1. Input volume is clustered to the n clusters where
n is the number of target class (tissues). The output
is clustered volume.
2. Under segmentation: If some clusters contain more
than one target class (tissue), user selects such clusters to be partitioned more; each user selected cluster
is re-clustered to two sub clusters. This process continues till user is satisfied. The output is clustered
volume without under segmentation.
3. Over segmentation: If several clusters correspond to
one target class (tissue), user selects clusters for each
target class. The output is final clustered volume.

i =

2 | Xi Yi |
| Xi | + | Yi |

rf pi =

| Yi | | Xi Yi |
| Xi |

rf ni =

| Xi | | Xi Yi |
| Xi |

(14)

where X i represents class i in ground truth and Y i

represents the same class in the segmentation result.
Each index for full segmentation results is the average
of that index for all classes.
3.1. Simulated volumes

The simulated MRI volumes are obtained from BrainWeb. A simulated data volume with T1-weighted
sequence, slice thickness of 1 mm and a volume size of
217 181 181 is used. Non-brain tissues are removed
prior to segmentation.
The number of tissue classes in the segmentation is
set to three: grey matter (GM), white matter (WM) and
cerebrospinal fluid (CSF). All pixels in the image are

3. Experimental Results and Discussion

The proposed extension of EM (EM-1) and the existing
neighbourhood-based extension of EM [19] (referred as
NWEM in this paper for clear understanding) are simulated and tested on the simulated volumes from BrainWeb
[30] and real volumes from Internet Brain Segmentation
Repository (IBSR) [31].

a)

b)

c)

Figure 1 clustering using user-interaction (a) A real brain volume, (b) its 4 clusters and (c) two sub clusters of Cluster 3.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

contributed in segmentation process but in evaluation

process, background pixels are ignored following previous works utilized in this paper. In the public databases which have been used in the paper and generally
in brain MRI volumes, background pixels have black
value. Therefore, cluster with lowest average grey value
is considered as background.
First, EM-1 and NWEM were applied to a slice of T1weighted brain image corrupted by different noise levels.
Figure 2 and Figure 3 show the segmentation results of
applying the afore-mentioned algorithms on a T1-weighted
normal brain slice in the presence of 9% and 7% rician
noise, respectively.
The segmentation results of white matter (WM), grey
matter (GM) and cerebrospinal fluid (CSF) are depicted
in. (a) is noisy image. (b) is ground-truth. (c) to (d) are the
segmentation results of NWEM and EM1, respectively.
From the above qualitative comparison, it was not difficult to find that NWEM was more influenced by the

a)

c)

Page 5 of 14

noise in comparison with EM1, in which fewer artefacts

were evident, resulting in clearer segmentation result.
Also, the proposed segmentation algorithm (EM-1)
and NWEM are applied to brain volume and average
similarity value is used to evaluate them. Figure 4 shows
the average similarity indexes r of mentioned algorithms
in different noise levels.
Figure 4 shows that EM-1 produces higher similarity
indexes and lower rfp and rfn, meaning that this algorithm produces more accurate segmentation results. The
similarity index of EM-1 decreases more slowly than
NWEM algorithm when noise level increases. In the
same time, the rfp and rfn of EM-1 increases faster than
NWEM algorithm.
Both algorithms give similar results, under 5% noise
level. However, for more than 5%, EM-1 exhibits much
better results than the NWEM algorithm. Incorporating
average of neighbourhood information, in clustering
process of NWEM, make this algorithm robust against

b)

d)

Figure 2 The segmentation results of applying EM1 and NWEM on a slice of image with 9% Rician noise. (a) Noisy image, (b) Groundtruth, Segmentation results of (c) NWEM and (d) EM1.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 6 of 14

Figure 3 The segmentation results of applying EM1 and NWEM on a slice of image with 7% Rician noise. (a) Noisy image, (b) Groundtruth, Segmentation results of (c) NWEM and (d) EM1.

noise but has blurring as side effect. It seems that with

increasing noise level more than 5% noise level; this
incorporation cannot overcome high level of noise.
Also the effect of different neighbourhood sizes on
performance of proposed segmentation algorithm (EM1) is investigated. Figure 5 shows the average similarity
index r of EM-1 for different neighbourhood sizes on
volume with 9% noise. Figure 5 shows that when the
neighbourhood size is increased, the similarity index of
EM-1 decreases sharply. This means blurring effect in
EM-1 depends on neighbourhood size.
The speed of EM1 and NWEM in segmenting a slice
was also investigated. Figure 6 represents the average
time required to segment a slice using the mentioned
algorithms. Figure 6 shows that EM1 is faster than
NWEM. The neighbourhood information in NWEM is
calculated in NWEM clustering iteration. Therefore, it is
time-consuming.
The proposed segmentation algorithm is also compared
with current extensions for EM. The average similarity

indexes r for proposed algorithm (EM-1) and several

current extensions for EM (DPM, rjMCMC, KVL and
MPM-MAP) are shown in Figure 7. Figure 7 shows that
EM-1 produces highest similarity indexes. The proposed
segmentation algorithm gives results comparable with
the best reported results, in low level of noise. However,
for noise levels more than 5%, EM-1 algorithm outperform other competing algorithms and this difference in
performance gets more in 9% noise level.
Also, EM-1 is compared with current existing neighbourhood based extensions for FCM. Figure 8 shows the
average similarity indexes r for EM-1 and FCM extensions
(FCM_S, FCM_S1, FCM_EN, FGFCM and FLICM) in different noise levels. At 3% noise level, the results for proposed segmentation algorithm and the best reported result
were close. Above 3% noise, EM-1 produces higher similarity index and were the most convincing in segmentation. The superiority of these algorithms increases with
increasing in noise level. FLICM shows worst performance
it seems it is not suit algorithm for brain volumes.

Similarity index

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

0.97
0.96
0.95
0.94
0.93
0.92
0.91
Noise0.9
level

3%

5%

7%

9%

0.9604

0.949

0.938

0.93

0.9624

0.9489

0.931

0.91

3%

5%

7%

9%

EM1

0.0356

0.051

0.0698

0.0684

NWEM

0.0322

0.0509

0.0733

0.0989

Ratio of false positive

EM1

Ratio of false negative

Page 7 of 14

NWEM
0.12
0.1
0.08
0.06
0.04
0.02
0
Noise level

0.09
0.08
0.07
0.06
0.05
0.04
0.03

Noise level

3%

5%

7%

9%

EM1

0.0386

0.0472

0.0497

0.0589

NWEM

0.0425

0.0546

0.0694

0.0922

Figure 4 The average similarity indices r, rfp and rfn for different noise levels.

In [25], the parameter-tied, constrained Gaussian mixture model (CGMM) is applied on image volume from
brainweb with different noise levels. Average similarity
index for different algorithms with variant noise levels
(3%, 5%, 7%, 9%) are: CGMM (0.93, 0.93, 0.92 and 0.895)
and KVL (0.925, 0.915, 0.895 and 0.865). The proposed
segmentation algorithm outperforms KVL and CGMM.
3.2. Real volumes

The superiority of our algorithm is also demonstrated on

real MRI volumes. The real MRI volumes are obtained
from the IBSR by the Centre for Morphometric Analysis
at Massachusetts General Hospital. 20 normal data

volume with T1-weighted sequence are used. First, proposed algorithm (EM-1) is applied to slices of a real MRI
volume with size 256*256*53. The average similarity
index r for volume image is 0.7986. Figure 9 shows the
similarity indexes of proposed algorithm (EM-1) for each
slice of MRI volume. In almost all slices, the proposed
algorithms exhibit better results for WM in compare to
results for GM. Better performance of proposed algorithms in WM is due to more simplicity and compactness
of WM in compare to GM.
EM-1 and NWEM are applied to all 20 normal real MRI
volumes and average similarity index r is used to compare
the segmentation results, quantitatively. Figure 10 shows

Similarity index

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

0.95
0.9
0.85
0.8
0.75
0.7

Neighbourhood size

EM1

Page 8 of 14

0.93

0.8674

0.7935

Figure 5 The average similarity index r for different neighbourhood sizes on simulated volume with 9% noise.

the average similarity index, rfp and rfn values of both

algorithms for all 20 normal volumes. Figure 10 shows
that EM-1 outperforms NWEM. EM-1 produces higher
average similarity indexes r and lower rfp and rfn.
The average similarity index values of proposed algorithm for 20 normal real MRI volumes and EM extensions (reported results in IBSR) are compared. Figure 11
shows the average similarity index values of different
algorithms for all 20 normal volumes. Figure 11 shows
that the similarity index for proposed segmentation
algorithms is higher than competing methods. It can be
seen clearly that proposed algorithm has a better performance over reported results, meaning that proposed
algorithm produces more accurate segmentation results
The proposed algorithms are also compared with
neighbourhood based extensions for FCM. Figure 12
shows the average similarity indexes r for proposed
algorithm and FCM extensions (FCM_S1, FCM_EN,
FGFCM) for all 20 normal volumes.
It can be seen clearly that proposed algorithm has a
better performance over FCM extension methods, and

produces more accurate segmentation results. FCM

extensions also incorporate neighbourhood information
in FCM clustering process, but, it seems that incorporating neighbourhood information improves EM more than
FCM method.
In [24], a novel trimmed minimum covariance determinant (TMCD) method an extension for Gaussian mixture model is applied on 20 normal image volumes from
IBSR. The average jaccard value was 0.6722. The average
jaccard values for EM-1 is: 0.695. The similarity index
for EM-1 is higher than reported result, meaning that
EM-1 produces more accurate segmentation results.
In [25], the parameter-tied, constrained Gaussian
mixture model (CGMM) is applied on 18 volumes
from 20 normal image volumes (except volume 4-8
and 202-3) in IBSR website. The CGMM results is
compared with reported results from the IBSR website,
as well as with the Marroquin algorithm [38]. Marroquins algorithm is an atlas-based Bayesian segmentation algorithm. The CGMM algorithm outperforms
other studied methods. Jacc similarity index CGMM

4.5
Average time (second)

4
3.5
3
2.5

EM1

NWEM

1.5
1
0.5
0

Algorithms

Figure 6 Average times required to segment a slice using the proposed algorithm (EM1) and NWEM.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 9 of 14

Similarity index

0.98
0.96
0.94
0.92
0.9
0.88
0.86
0.84

Noise level

3%

5%

7%

9%

EM1

0.9604

0.949

0.938

0.93

DPM

0.942

0.927

0.902

0.894

rjMCMC

0.92

0.897

0.882

0.875

KVL

0.922

0.923

0.901

0.874

MPM-MAP

0.955

0.942

0.905

0.875

Figure 7 The average similarity indices r for different noise levels.

Similarity index

0.98
0.96
0.94
0.92
0.9
0.88
0.86
0.84
0.82
0.8
0.78

3%

5%

7%

9%

Noise level
EM-1

0.9604

0.9485

0.938

0.93

FCM-S

0.9505

0.9307

0.914

0.901

FCM-EN

0.94

0.933

0.9298

0.918

FGFCM

0.9505

0.9308

0.915

0.89

NOnlocalFCM

0.9605

0.9402

0.9203

0.9106

FLICM

0.8677

0.8492

0.8223

0.7854

Figure 8 The average similarity indices r for EM-1 and FCM extensions in different noise level.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 10 of 14

Ratio of false negative

Ratio of false positive

Average similarity
index

Figure 9 The similarity index of proposed algorithm when applied for real volume.

0.85
0.8
0.75
0.7
0.65

Algorithms

EM-1

0.8211

NWEM

0.7315

0.18
0.16
0.14
0.12
0.1

Algorithms

EM-1

0.1373

NWEM

0.1717

0.35
0.3
0.25
0.2
0.15
0.1

Algorithms

EM-1

0.2069

NWEM

0.3138

Figure 10 The average similarity index, rfp and rfn of proposed algorithm when applied on 20 real volumes.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 11 of 14

Similarity index

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

Algorithms

EM-1

0.6965

adaptive MAP

0.5655

biased MAP

0.56

fuzzy c-means

0.52

Maximum Aposteriori
Probability (MAP)

0.552

Maximum-Likelihood

0.543

tree-structure k-means

0.524

Manual (4 brains
averaged over 2 experts)

0.854

Figure 11 The average similarity index of different algorithms when applied on 20 real volumes.

was: 0.67. The average jaccard values for EM-1 is:

0.6971. EM-1 outperforms the best reported result
which is for CGMM.
In [26], a nonparametric Bayesian model, known as
Dirichlet process mixture model (DPMM) is applied on 13
volumes (1_24, 2_4, 5_8, 6_10, 7_8, 11_3, 12_3, 13_3,
15_3, 16_3, 100_23, 110_3 112_2) from the 20 normal T1weighted brain image volumes from IBSR. The similarity
index for DPMM is higher than competing methods. Dice
similarity index for DPMM was: 0.7071. The proposed
algorithms are applied on the same volumes. The average
Dic value for EM-1 is: 0.8219. The similarity index for proposed method is higher than the best reported result
which is for DPMM, meaning that proposed method are
the most convincing in segmentation.

In [27], local cooperative unified segmentation

(LOCUS) approach which is based on distributed local
MRF models for brain segmentation is applied on the
20 normal T1-weighted brain image volumes from IBSR.
LOCUS-T is compared with published results for SPM5
and FAST. Dic similarity index for different methods
are: LOCUS-T = 0.765, SPM5 = 0.81, FAST = 0.765.
The average Dic value for EM-1 is: 0.8211. EM-1 outperforms the best reported result which is for SPM5.
Also, improvement of segmentation result using userinteraction is investigated. Proposed algorithms and the
same algorithm with user-interaction are applied to all 20
normal real MRI volumes and similarity index r is used
to compare the segmentation results, quantitatively. The
average similarity index values of both algorithms for

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 12 of 14

0.84

Similarity index

0.82
0.8
0.78
0.76
0.74
0.72
0.7

Algorithms

EM-1

0.8211

FCM-S

0.7517

FCM-EN

0.7581

FGFCM

0.7597

Figure 12 The similarity index of proposed algorithm and neighbourhood based FCM extensions when applied on 20 real volumes.

different volume were presented in Figure 13 and Table

1. Figure 13 shows that user-interaction improves performance of proposed algorithm and increases similarity
indexes r in all image volumes.

4. Conclusion
In this paper, an extension of EM has been introduced. In
order to overcome the problem of standard EM in the presence of noise, the introduced algorithms are formulated
by modifying the equations of the standard EM algorithm
which allow the neighbourhood pixels to be incorporated
in the labelling of a pixel. Introduced algorithm is tested

on simulated MRI volumes, with different noise levels and

real volumes. The performance of the existing neighbourhood based EM and FCM algorithms and proposed algorithm are compared qualitatively.
The similarity index, r is used to evaluate different
algorithms. Experiments demonstrate the effectiveness
of the proposed algorithm in compare to other existing
algorithms on various noise levels in terms of similarity
index, r.
In future, we consider doing research on other kinds
of segmentation methods to improve their functionalities. Also, we will analyse the effects of different

0.9
Similarity index

0.85
0.8
0.75

EM-1 with user-interaction

0.7

EM-1

0.65
1_24
2_4
5_8
4_8
6_10
7_8
8_4
11_3
12_3
13_3
15_3
16_3
17_3
100_23
110_3
111_2
112_2
191_3
202_3
205_3

0.6

Volumes
Figure 13 The similarity index of different algorithms when applied on 20 real volumes.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 13 of 14

Table 1 The similarity index of different algorithms when applied on 20 real volumes
Volumes Algorithms

EM-1 with user-interaction

EM-1

1_24

0.8678

0.8454

2_4

0.8136

0.7899

5_8

0.8462

0.8076

4_8

0.8091

0.775

6_10

0.8518

0.8091

7_8

0.8613

0.8386

8_4

0.8501

0.8269

11_3

0.8675

0.8471

12_3

0.8498

0.8263

13_3

0.8458

0.8243

15_3

0.8169

0.7691

16_3

0.8256

0.7887

17_3

0.8582

0.811

100_23

0.8649

0.8506

110_3

0.8555

0.8407

111_2

0.8452

0.8095

112_2

0.857

0.8475

191_3

0.8566

0.8322

202_3

0.8555

0.8453

205_3

0.8611

0.8377

clustering methods in segmentation of medical images

for the diagnosis of abnormal or various important matters in medical images.

9.

10.

Authors contributions
MA performed all works for this paper.

11.

Competing interests
The authors declare that they have no competing interests.

12.
13.

Received: 10 June 2011 Accepted: 26 October 2011

Published: 26 October 2011
14.
References
1. Chang PL, Teng WG: Exploiting the self-organizing map for medical
image segmentation. Twentieth IEEE International Symposium on
Computer-Based Medical Systems 2007, 281-288.
2. Jan J: Medical image processing, reconstruction, and restoration: concepts and
methods: CRC 2006.
3. Tian D, Fan L: A Brain MR Images Segmentation Method Based on SOM
Neural Network. The 1st International Conference on Bioinformatics and
Biomedical Engineering 2007, 686-689.
4. Jiang Y, Meng J, Babyn P: X-ray image segmentation using active
contour model with global constraints. 2007, 240-245.
5. Balafar MA: New spatial based MRI image de-noising algorithm. Artifitial
Intelligence Review 2011, 1-11.
6. Balafar MA, Ramli AR, Mashohor S: A New Method for MR Grayscale
Inhomogeneity Correction. Artificial Intelligence Review, Springer 2010,
34:195-204.
7. Balafar MA, Ramli AR, Saripan MI, Mashohor S: Review of brain MRI image
segmentation methods. Artificial lntelligence Review 2010, 33:261-274.
8. Balafar MA, Ramli AR, Mashohor S: Compare different spatial based Fuzzy
C-Mean (FCM) extensions for MRI Image Segmentation. ICCAE 2010,
609-611.

15.

16.

17.

18.

19.

20.

Balafar MA, Ramli AR, Mashohor S: Edge-preserving Clustering Algorithms

and Their Application for MRI Image Segmentation. International
MultiConference of Engineers and Computer Scientists 2010, 17-19.
Balafar MA, Ramli A-R, Mashohor S: Medical Brain magnetic resonance
image segmentation using novel improvement for expectation
maximizing. Neurosciences 2011, 16:242-247.
Balafar MA, Ramli AR, saripan MI, Mashohor S, Mahmud R: Improved Fast
Fuzzy C-Mean and its Application in Medical Image Segmentation.
Journal of Circuits, Systems, and Computers 2010, 19:203-214.
Zou K, Wang Z, Hu M: An new initialization method for fuzzy c-means
algorithm. Fuzzy Optimization and Decision Making 2008, 7:409-416.
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 2006, 30:9-15.
He R, Datta S, Sajja BR, Narayana PA: Generalized fuzzy clustering for
segmentation of multi-spectral magnetic resonance images.
Computerized Medical Imaging and Graphics 2008, 32:353-366.
Balafar MA, Ramli AR, Saripan MI, Mashohor S, Mahmud R: Medical Image
Segmentation Using Fuzzy C-Mean (FCM) and User Specified Data.
Journal of Circuits, Systems, and Computers 2010, 19:1-14.
Balafar MA, Ramli AR, Saripan MI, Mahmud R, Mashohor S: Medical image
segmentation using fuzzy C-mean (FCM) and dominant grey levels of
image. Visual information engineering conference 2008, 314-317.
Balafar MA, Ramli AR, Saripan MI, Mahmud R, Mashohor S: MRI
segmentation of medical images using FCM with initialized class centers
via genetic algorithm. International symposium on information technology
2008, 1-4.
Zhang Y, Brady M, Smith S: Segmentation of brain MR images through a
hidden Markov random field model and the expectation-maximization
algorithm. IEEE Transactions on Medical Imaging 2001, 20:45-57.
Tanga H, Dillensegerb J, Baoa XD, Luoa LM: A Vectorial Image Soft
Segmentation Method Based on Neighborhood Weighted Gaussian
Mixture Model. Computerized Medical Imaging Graphics 2009, 33:644-650.
Silva ARFd: Bayesian mixture models of variable dimension for image
segmentation. computer methods and programs in biomedicine 2009,
94:1-14.

Balafar Diagnostic Pathology 2011, 6:103

http://www.diagnosticpathology.org/content/6/1/103

Page 14 of 14

21. Leemput FMKV, Vandermeulen D, Suetens P: Automated model-based

tissue classification of MR images of the brain. IEEE Transactions on
Medical Imaging 1999, 18:897-908.
22. Marroquin BCVJL, Botello S, Calderon F, Fernandez-Bouzas A: An accurate
and efficient Bayesian method for automatic segmentation of brain
MRI. IEEE Transactions on Medical Imaging 2002, 21:934-945.
23. Ballester MG, Zisserman A, Brady M: Estimation of the partial volume
effect in MRI. Medical Image Analysis 2002, 6:389-405.
24. Tohka J, Zijdenbos A, Evans A: Fast and robust parameter estimation for
statistical partial volume models in brain MRI. NeuroImage 2004,
23:84-97.
25. Greenspan H, Ruf A, Goldberger J: Constrained Gaussian mixture model
framework for automatic segmentation of MR brain images. IEEE
transactions on medical imaging 2006, 25:1233-1245.
26. Silva ARFd: A Dirichlet process mixture model for brain MRI tissue
classification. Medical Image Analysis 2007, 11:169-182.
27. Scherrer B, Forbes F, Garbay C, Dojat M: Distributed Local MRF Models for
Tissue and Structure Brain Segmentation. IEEE TRANSACTIONS ON
MEDICAL IMAGING 2009, 28:1278-1295.
28. Gudbjartsson H, Patz S: The Rician distribution of noisy MRI data.
Magnetic resonance in medicine: official journal of the Society of Magnetic
Resonance in Medicine/Society of Magnetic Resonance in Medicine 1995,
34:910.
29. Balafar MA, Ramli AR, Saripan MI, Mahmud R, Mashohor S: Medical image
segmentation using fuzzy C-mean (FCM), Bayesian method and user
interaction. International conference on wavelet analysis and pattern
recognition 2008, 68-73.
30. , BrainWeb [Online]: http://mouldy.bic.mni.mcgill.ca/brainweb/, Last
accessed october, 2010..
31. , IBSR. Available: http://www.cma.mgh.harvard.edu/ibsr/, Last accessed
october, 2010..
32. Ahmed MN, Yamany SM, Mohamed N, Farag AA, Moriarty T: A modified
fuzzy c-means algorithm for bias field estimation and segmentation of
MRI data. IEEE Transactions on Medical Imaging 2002, 21:193-199.
33. Szilgyi L, Beny Z, Szilgyii SM, Adam HS: MR brain image segmentation
using an enhanced fuzzy c-means algorithm. 25th Annual International
Conference of IEEE EMBS 2003, 17-21.
34. Cai W, Chen S, Zhang D: Fast and robust fuzzy c-means clustering
algorithms incorporating local information for image segmentation.
Pattern Recognition 2007, 40:825-838.
35. Krinidis S, Chatzis V: A Robust Fuzzy Local Information C-Means
Clustering Algorithm. IEEE Transactions on Image Processing 2010,
19:1328-1337.
36. Wang J, Kong J, Lub Y, Qi M, Zhang B: A modified FCM algorithm for
MRI brain image segmentation using both local and non-local spatial
constraints. Computerized Medical Imaging and Graphics 2008, 32:685-98.
37. Zijdenbos AP, Dawant BM: Brain segmentation and white matter lesion
detection in MR images. Crit Rev Biomed Eng 1994, 22:401-465.
38. Marroquin J, Vemuri B, Botello S, Calderon F, Fernandez-Bouzas A: An
accurate and efficient Bayesian method for automatic segmentation of
brain MRI. IEEE Transactions on Medical Imaging 2002, 21:934-945.
doi:10.1186/1746-1596-6-103
Cite this article as: Balafar: Spatial based Expectation Maximizing (EM).
Diagnostic Pathology 2011 6:103.

Submit your next manuscript to BioMed Central

and take full advantage of:
Convenient online submission
Thorough peer review
No space constraints or color figure charges
Immediate publication on acceptance
Inclusion in PubMed, CAS, Scopus and Google Scholar
Research which is freely available for redistribution
Submit your manuscript at
www.biomedcentral.com/submit