Page 1

The Open Medical Informatics Journal, 2010, 4, 41-49 41

1874-4311/10 2010 Bentham Open

Open Access

Segmentation of Fluorescence Microscopy Cell Images Using Unsupervised

Mining

Xian Du1 and Sumeet Dua*,1,2

1Data Mining Research Laboratory, Department of Computer Science, College of Engineering and Science, Louisiana

Tech University, Ruston, LA, USA

2School of Medicine, Louisiana State Health Sciences Center, New Orleans, LA, USA

Abstract: The accurate measurement of cell and nuclei contours are critical for the sensitive and specific detection of

changes in normal cells in several medical informatics disciplines. Within microscopy, this task is facilitated using

fluorescence cell stains, and segmentation is often the first step in such approaches. Due to the complex nature of cell

issues and problems inherent to microscopy, unsupervised mining approaches of clustering can be incorporated in the

segmentation of cells. In this study, we have developed and evaluated the performance of multiple unsupervised data

mining techniques in cell image segmentation. We adapt four distinctive, yet complementary, methods for unsupervised

learning, including those based on k-means clustering, EM, Otsu’s threshold, and GMAC. Validation measures are

defined, and the performance of the techniques is evaluated both quantitatively and qualitatively using synthetic and

recently published real data. Experimental results demonstrate that k-means, Otsu’s threshold, and GMAC perform

similarly, and have more precise segmentation results than EM. We report that EM has higher recall values and lower

precision results from under-segmentation due to its Gaussian model assumption. We also demonstrate that these methods

need spatial information to segment complex real cell images with a high degree of efficacy, as expected in many medical

informatics applications.

Keywords: Fluorescence microscope cell image, segmentation, K-means clustering, EM, threshold, GMAC.

1. INTRODUCTION

medical informatics disciplines, including but not limited to,

cancer informatics, neuro-informatics, and clinical decision

support in ophthalmology. While fluorescence microscopes

permit the collection of large, high-dimensional cell image

datasets, their manual processing is inefficient, irrepro-

ducible, time-consuming, and error-prone, prompting the

design and development of automated, efficient, and robust

processing to allow analysis for high-throughput applicat-

ions. The sensitive and specific detection of pathological

changes in cells requires the accurate measurement of

geometric parameters. Previous research has shown that

geometric features, such as shape and area, indicate cell

morphological changes during apoptosis [1]. As a precursor

to geometric analysis, segmentation is often required in the

first processing step. Cell image segmentation is challenging

due to the complex morphological cells, illuminant reflect-

ion, and inherent microscopy noises. The characteristic

problems include poor contrast between cell gray levels and

background, a high number of occluding cells in a single

view, and excess homogeneity in cell images due to irregular

staining among cells and tissues.

Microscopic imaging is nearly ubiquitous in several

local image information, including edge or gradient, level set

Typically, image segmentation algorithms are based on

*Address correspondence to this author at the Data Mining Research

Laboratory, Department of Computer Science, College of Engineering and

Science, Louisiana Tech University, Ruston, LA 71272, USA; Tel: 318-

257-2830; Fax: 318-257-4922; E-mail: sdua@latech.edu

[2], histogram [3], clusters [4], and prior knowledge [5].

These segmentation methods have been broadly implemen-

ted in medical imaging applications [6]. The current

segmentation algorithms used in cell images include seeded

watershed [7], Voronoi-based algorithm [8], histogram-

based clustering [9] or threshold [10] and active contour

[11]. Watershed algorithms can split the connected cells but

can lead to over-segmentation. Histogram-based image

segmentation is unparametric and based on unsupervised

clustering. The histogram is used to approximate the

probability density distribution of the image intensity. Pixels

in one image are partitioned into several non-overlapping

intensity regions. K-means and EM are extensions of

histogram segmentation. In EM [9], the distribution of image

intensity is modeled as a random variable, which is

approximated by a mixture Gaussian model. Due to the lack

of intensity distribution information in an image, the EM

model can lead to significant bias. of the EM model is

computationally efficient and easy to implement, but

performs poorly in finding the optimal threshold between

clusters in the histogram. Otsu’s optimal threshold is

obtained by minimizing intra-class variance and has been

applied in nucleus segmentation [12]. Level set and active

contour are applied with arbitrary interaction energy in order

to split the connected cells in [11]. This method is not

meaningful for isolated cells and makes the cell

segmentation dependent on cell sizes. In [8], cells are

segmented according to the defined metric, the Voronoi

distance between pixels and seeds. This metric includes the

information from image edges and inter-pixel distance

within the image. The parametric active contour and

Page 2

42 The Open Medical Informatics Journal, 2010, Volume 4 Du and Dua

repulsive force are incorporated in [13]. However, this

metric is not suitable for the segmentation of a large number

of cells in one image.

nuclei and cell image segmentation due to the inherent

coherent detection and decomposition challenges in the

detection and separation of segments. However, it is difficult

to select a robust and reproducible method due to the lack of

the comparative evaluation of those algorithms. This

problem arises partially due to the lack of benchmark data or

because of manually outlined ground truth. This paucity of

performance evaluation elevates the difficulty for medical

scientists to select a suitable segmentation method in medical

image applications. Sometimes, methods are selected based

on intuition and experience; e.g., Otsu’s threshold is used

broadly in nuclei image segmentation. Moreover, no broadly

acceptable method can address the nuclei and cell image

segmentation problems in a diverse range of applications

accurately and robustly. Recently, several synthetic (e.g.

[14]) and benchmark cellular image data (e.g. [15]) have

been made publicly available.

Unsupervised learning can be adapted and developed for

several unsupervised data mining techniques in cell image

segmentation. We adapt four distinctive, yet complementary,

methods for unsupervised learning, including those based on

k-means clustering, EM, Otsu’s threshold, and GMAC.

Validation measures are defined to compare and contrast the

performance of these methods using publicly available data.

It should be noted that the segmentation algorithms are

typical representatives of methods based on histogram,

model, threshold, and active contour. We only focus on

segmentation methods using low-level image information,

such as pixel intensity and image gradient. GMAC

represents both the snake and level set technologies [14].

The results presented in this paper can guide domain users to

select suitable segmentation methods in medical imaging

applications.

In this paper, we present and evaluate the performance of

2. UNSUPERVISED MINING METHODS FOR IMAGE

SEGMENTATION

where each pixel can take L possible grayscale-level values

in the range [0, L?1]. Let h(x) be the normalized histogram

of the image I.

Let us consider an image I of size r = M ? N pixels,

2.1. Notation

xi Intensity value of pixel i

h(x) Histogram of the image I, x ? 0,L ?1

[]

r

TrI ( )

pjxi;?j

Image size in terms of pixel numbers

Transformation function of image I

()

j-th probability density function with

parameter set ?j

Mean of cluster j

μj

?j

?within

?between

Variance of cluster j

2

2

Within-class variance,

Between-class variance.

?iT ( )i=1,2

Probabilities

separated by threshold T

of the two clusters

f x ( )

Image expressed with spatial term x,

which refers to pixel location

? (in GMAC)

Scalar that controls the balance between

regularization and data

2.2. K-Means Clustering

find the optimal threshold, such that the image feature values

of pixels on one side of the threshold are closer to their

feature values’ mean than the distance between those feature

values and the means on the other side of the threshold. This

method is performed using the histogram of image intensity.

We assume that the image intensities compose a vector space

and try to find natural clustering in it. The pixels are

clustered around centroid ci, which are obtained by

minimizing the objective function

(

We use K-means clustering for image segmentation to

ci:= argmin dist xi? μj

)

(). (1)

follows,

The centroid for each cluster is iteratively obtained as

μi:=

ci= j

ci= j

{

{}xi

}

i=1

r

?

?

i=1

r

, (2)

where r is the image size in terms of pixel number, i iterates

over all intensities, j iterates over all centroids, and μi are the

centroid intensities. Using intensity value directly in

microscopic cell image segmentation will not lead to the

desired segmentation result due to the dynamic ranges,

which vary in images. We propose a gray-level

transformation function in the form TrI ( )= Ir for the above

algorithm to implement k-means segmentation in cell image

I, where ? is a positive constancy.

2.3. Expectation Maximization Method

that an image consists of a number of gray-level regions,

which can be described by parametric data models. When the

histogram of the gray levels is regarded as an estimate of the

probability density function, the parameters of the function

can be estimated for each gray-level region using the

histogram. The objective of the EM algorithm is to find the

maximum likelihood estimates of the parameters in the

function. Correspondingly, EM consists of two steps:

expectation and maximization. Using the same notations in

Section 2.1, the mixture of probability density functions is as

follows,

(

j=1

The Expectation Maximization (EM) algorithm assumes

p xi ( )=?jpjxi;?j

?

)

K

. (3)

In the above, ?jis the proportion of the j-th density

function in the mixture model, and

?jpjxi;?j

?

()

j=1

K

is the j-th

Page 3

Segmentation of Fluorescence Microscopy Cell Images Using Unsupervised Mining The Open Medical Informatics Journal, 2010, Volume 4 43

density function with parameter set ?j. The Gaussian

mixture model (GMM) is the most employed in practice, and

has two parameters, mean μjand covariance ?j, such that

?j= μj,?j

?j

is the

parameters?j= μj,?j

?j

framework follows,

(). We consider GMM in our research. If we

that

(

t+1can be obtained iteratively. The EM algorithm

assume

t estimated value of

), obtained at the t-th step, then

?j

t+1=1

r

?ij

t

i=1

r

?

, (4)

μj

t+1=

?ij

txi

i=1

?

r

?

?ij

t

i=1

r

, (5)

?j

t+1=

?ij

t

xi? μj

t

() xi? μj

t

()

T

?

??

?

??

i=1

r

?

?ij

t

i=1

r

?

, (6)

?ij

t=

?j

?

tp xi;μj

(

t,?j

t

()

?j

tp xi;μj

t,?j

t

)

j=1

K

. (7)

density function are updated according to the weighted

average of the pixel values where the weights are obtained

from the E step for this partition. The EM cycle starts at an

initial setting of ?j

using Equations ((4)-(7)) iteratively. The EM algorithm

converges until its estimated parameters cannot change.

Then, the final parameters,?j

image segmentation by labeling pixels using Maximum

Likelihood (ML). Pixel xi is labeled using the following

function,

(

?j

These equations state that the estimated parameters of the

0= μj

0,?j

0

() and updates the parameters

EM= μj

EM,?j

EM

(), are applied in

argmax

j

exp ?0.5 xi? μj

EM

)?j

EMxi? μj

EM

()

()

EM?0.5

.

(8)

2.4. Threshold-Based Segmentation

image into a number of meaningful regions through the

selected threshold values. If the image is a grey image,

thresholds are integers in the range of [0, L-1], where L-1 is

the maximum intensity value. Normally, this method is used

to segment an image into two regions: background and

object, with one threshold. The following is the equation for

threshold segmentation:

Threshold segmentation is a method that separates an

IBx,y

()=

1, ifI x,y

(

(

)>T

)?T.

0,if I x,y

?

?

??

?

. (9)

The most famous threshold method was proposed by Otsu in

[12]. The Otsu’s method finds the optimal threshold T

among all the intensity values from 0 to L-1 and chooses the

value that produces the minimum within-class variance

?within

as the optimal threshold value. Consequently, the

optimal value of TOpt is obtained through the following

optimal computation,

(

In the above equation, IB is the segmentation resultant.

2

?within

2

Topt

)= min

0?T?L?1?within

2

T ( )

????. (10)

parts:?2= ?within

??

Therefore, the optimal value of T can also be obtained

through the following alternative optimization process:

(

In the whole image, variances ?2are made up of two

2

T ( )+?between

0?T?L?1?within

2

T ( ) . Otsu shows

??

that

??.

min

2

T ( )

?? is the same as max

0?T?L?1?between

2

T ( )

?between

2

Topt

)= max

0?T?L?1?between

2

T ( )

????. (11)

value for simple calculation. Theoretically, ?between

expressed in the following,

(

Equation (11) is often used to find the optimal threshold

2

T ( )is

?between

2

Topt

)=?1T ( )?2T ( ) μ1T ( )? μ2T ( )

()

2

(12)

where ?iT ( )=

h i ( )

i=1

t

?

are the probabilities of the two

clusters separated by threshold T, and μiT ( )i=1,2are the

cluster means. ?iT ( )i=1,2and μiT ( )i=1,2can be estimated

using histogram h(x) as follows,

?1T ( )=

h i ( )

i=1

T

?

(13)

?2T ( )=

h i ( )

i=T+1

L?1

?

, (14)

μ1T ( )=

i?h i ( )

i=0

T

?

?1

, (15)

μ2T ( )=

i?h i ( )

i=T+1

L?1

?

?2

. (16)

threshold T is exhaustively searched among [0, L-1] to meet

the objective according to Equation (11).

Using the above Equations (12)-(16), the optimal

2.5. Global Minimization of the Active Contour Model

(GMAC)

model (GMAC) [16] to analyze the implementation of active

contour in cell-image segmentation. This method has a

simple initialization and fast computation, and it can avoid

being stuck at an undesired local minima. GMAC is based

on Mumford and Shah’s (MS) function and the Chan and

Vese’s model of active contours without edges (ACWE)

We choose the global minimization of the active contour

Page 4

44 The Open Medical Informatics Journal, 2010, Volume 4 Du and Dua

[17]. GMAC improves ACWE by using weighted total

variation and dual formulation of the TV form, which

preserves the advantage of ACWE. We define GMAC and

related concepts below.

min

μ,?EGMACμ,?

(

():= TVgμ ( )+1

)μ

2?

μ ?vL2

2

+? r1x,c1,c2

??

+?? v ( )dx

, (17)

where r1x,c1,c2

()=

c1? f x ( )

()

2? c2? f x ( )

()

2

()dx , f x ( )is

the given image, and c1 and c2 are constants calculated for

partitioning in iteration; e.g., if μ*= argminE2μ,v,c1,c2

c1 and c2 are the means of pixels in two partitions and can be

obtained using equations, ? > 0 is chosen small enough,

? > 0 is a parameter controlling scale related to the scale of

observation of solution, and ? is constant.

?

where g(x) is an edge indication function which gives a link

between snake model and region terms. The minimization

Equation (17) is solved using the following equations

iteratively until convergence:

[],

TVgμ ( )=

g x ( ) ?μ dx

(18)

c1=

f x ( )

r?

v x ( )dx

v x ( )dx

r?

, (19)

c2=

f x ( ) 1?v x ( )

(

(

)

r?

dx

1?v x ( )

)dx

r?

, (20)

pn+1=

pn+?t? divpn? f ?v

(

1+

g x ( )? divpn? f ?v

()/?

)/?

)

?t

(

()

, (21)

μ = v?? ?divp, (22)

v x ( )= min max μ x ( )???r1x,c1,c2

(),0

{},1

{} (23)

In Equation (21),?t is the time step.

3. EXPERIMENTAL RESULTS

the segmentation of three types of fluorescent cellular

images: synthetic cell images, nuclei images with ground

truth, and brain cell microscopic images. The first two types

of image data are used to evaluate the quantitative

performance of the four segmentation methods and to

compare the results to the ground truth. The brain cell

images are segmented with qualitative performance analysis

due to the lack of ground truth.

In this section, we present the experimental results from

3.1. Quantitative Measure

the quantitative measures in pixel level. These measures are

standard techniques used to evaluate the quality of the

segmentation results against the ground truth. These

We use the traditional precision, recall, and F-score as

measures quantify discrepancy between segmentation results

and binary ground truth mask as follows:

()

#SR

()

#GT

F ? score =2? precision?recall

precision+recall

precision =# SR?GT

, (24)

recall =# SR?GT

, (25)

()

, (26)

where SR is the segmentation result and GT is the ground

truth of images. The symbol ‘#’ refers to the pixel numbers

in the sets.

3.1.1. Segmentation of Synthetic Data

ground truth are simulated by P. Ruusuvuori in [18]. We

select the second benchmark set which consists of multi-

channel cell images because we do not have suitable real cell

images with ground truth for evaluation. In this set, nuclei,

cytoplasm, and subcellular components have been simulated

by tuning parameters such as size, location, randomness of

shape, and other background or fluorescence parameters (see

details in [18]). The image sets are divided into two subsets:

high quality and low quality (examples shown in Fig. 1),

each consisting of 20 cell images. The second set has

overlapping cells and a noisy background. Each image

contains 50 cells. As each simulated image has a

corresponding binary mask as ground truth, binary

operations can easily calculate the quantitative measure

defined above.

Benchmark sets of synthetic cell population images with

a) b)

c) d)

Fig. (1). Synthetic cell images a) (low quality) with noisy

background and overlapping cells, b) (high quality) without noise

in background and overlapping cells, c) ground truth of image a, d)

ground truth of image b.

for the low quality synthetic image data in Fig. (1a).

Fig. (2) shows the segmentation result of four methods

Page 5

Segmentation of Fluorescence Microscopy Cell Images Using Unsupervised Mining The Open Medical Informatics Journal, 2010, Volume 4 45

Segmented images in Fig. (2) are compared and evaluated

using the ground truth image in Fig. (1c). Fig. (3) shows the

segmentation result of four methods for the high quality

synthetic image data in Fig. (1b). Segmented images in Fig.

(3) are compared and evaluated using the ground truth image

in Fig. (1d).

a) b)

c) d)

Fig. (2). Segmentation result for synthetic cell images of low

quality in Fig. (1a). a) K-means result, b) EM result, c) Otsu’s

result, d) GMAC result.

a) b)

c) d)

Fig. (3). Segmentation result for synthetic cell image of high

quality in Fig. (1b). a) K-means result, b) EM result, c) Otsu’s

result, d) GMAC result.

the segmentation results using subcellular images with low

quality. Figs. (7-9) and Table 2 are the quality measure

Figs. (4-6) and Table 1 are the quality measure values for

values for the segmentation results using subcellular images

with high quality.

images, with noisier backgrounds and overlapping cells,

have worse results than those in high quality images. K-

means, Otsu’s threshold and GMAC obtain similar

segmentation quality in both sets of images, measured by F-

score, precision, and recall. Their performance is more

robust against noises than EM. Moreover, the EM algorithm

has lower precision, while keeping much higher recall

values, especially for cell images with noisy backgrounds.

To further understand these phenomena, real nucleus images

are segmented in the next section.

We observe that the segmentation results of lower quality

Table 1. Average Measures of the Segmentation Methods

Applied on Low Quality Synthetic Cell Images

F-Score Precision Recall

K-Means 0.9350 0.9530 0.9180

EM 0.5331 0.3821 0.9915

Otsu’s 0.9269 0.9295 0.9259

GMAC 0.9445 0.9781 0.9133

Table 2. Average Measures of the Segmentation Methods

Applied on High Quality Synthetic Cell Images

F-Score Precision Recall

K-Means 0.9745 0.9726 0.9765

EM 0.9040 0.8267 0.9986

Otsu’s 0.9738 0.9798 0.9679

GMAC 0.9703 0.9874 0.9538

Fig. (4). F-score of the four methods applied on low quality

simulated cell images.

3.1.2. Segmentation of Nucleus Images

in the CellProfiler project [8, 15]. We use these images to

evaluate the segmentation algorithms quantitatively. We

obtain similar results, as shown in Figs. (10-13), as those we

obtained in Section 3.1.1. We observe in Table 3 that EM

Sixteen nucleus images were hand-outlined by an expert

0.5

0.6

0.7

0.8

0.9

1

135791113151719

Image Number

F-Score

K-means

Otsu's

EM

GMAC

Page 6

46 The Open Medical Informatics Journal, 2010, Volume 4 Du and Dua

maintains higher recall and lower precision values, even if its

F-score values are as high as the other segmentation methods

in several images. From Fig. (10d), we can see that the EM

under-segments nucleus images strongly, which induces the

high recall values. This under-segmentation is due to the

presumed dual Gaussian mixture models in the calculation of

EM. One model represents background, and the other refers

to objects. When objects have much smaller grayness

regions than background (as shown in Fig. 10a), the dual

Gaussian mixture model leads to under-segmentation.

Fig. (5). Precision of the four methods applied on low quality

simulated cell images.

Fig. (6). Recall of the four methods applied on low quality

simulated cell images.

Fig. (7). F-score of the four methods applied on high quality

simulated cell images.

Fig. (8). Precision of the four methods applied on high quality

simulated cell images.

Fig. (9). Recall of the four methods applied on high quality

simulated cell images.

well for nucleus segmentations, due to its fastness and

simplicity in application, it cannot be proven the best

segmentation method for nucleus images. As shown in

Section 3.1.1, Otsu’s method shows stable precision and

recall values even when it encounters arbitrarily defined

noises. However, in the experiment using real nucleus

images, the Otsu’s method recall value is significantly lower

than its precision values, which means it has over-segmented

the image.

Otsu’s method also has drawbacks. Although it performs

in our experiments. GMAC depends on both image intensity

distribution information (region) and gradient (edge)

information. When the contrast between background and

cells becomes light, and cells are hidden by noises, the

combination of gradient and intensity information records

better information than intensity alone does, e.g. in Otsu’s.

In the k-means method, we choose k=2 to cluster some

objects into one group and other segments into a background

group. K-means performs the best in almost all experiments.

Its good performance is due to the application of power

function for the compensation of intensity transformation

brought in by the microscopic device. In this research, we

assume this power function is known, and we obtain it by

choosing the optimal k-means result (smallest error between

GMAC is more robust and stable than the Otsu’s method

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1.1

1357

Image Number

91113151719

Precision

K-means

Otsu's

EM

GMAC

0.85

0.87

0.89

0.91

0.93

0.95

0.97

0.99

1.01

135791113151719

Image Number

Recall

K-means

Otsu's

EM

GMAC

0.85

0.87

0.89

0.91

0.93

0.95

0.97

0.99

1357

Image Number

91113151719

F-Score

K-means

Otsu's

EM

GMAC

0.75

0.8

0.85

0.9

0.95

1

1.05

1357

Image Number

91113151719

Precision

K-means

Otsu's

EM

GMAC

0.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1

1.01

1357

Image Number

91113151719

Recall

K-means

Otsu's

EM

GMAC

Page 7

Segmentation of Fluorescence Microscopy Cell Images Using Unsupervised Mining The Open Medical Informatics Journal, 2010, Volume 4 47

k-means segmentation result and ground truth). It

demonstrates that the k-means method can obtain robust and

precise segmentation results with the aid of power function.

a) b)

c) d)

e) f)

Fig. (10). Segmentation of nucleus images: a) Nucleus images, b)

Ground truth, c) K-means result, d) EM result, e) Otsu’s result, and

f) GMAC result.

Table 3. Average Quality Measures of the Segmentation

Methods on Nucleus Images

F-Score Precision Recall

K-Means 0.8714 0.8766 0.8668

EM 0.7473 0.6131 0.9664

Otsu’s 0.7976 0.9475 0.6910

GMAC 0.7880 0.8880 0.7148

3.2. Quality Measure of Segmentation of Brian Cell

Images

using a computer controlled Microscope (Leica DMI 6000

Digital). The cell images are of a normal healthy astrocytes

In this evaluative study, brain cell images were captured

cell, which has been stained with Calcein AM, a vital dye

that stains only living cells. The test images are

pixels with 8-bit gray-levels. As no manual outlining has

been performed on the images, the performance of

segmentation methods is qualitatively evaluated.

↔

10401392

Fig. (11). F-score of the four methods applied on nucleus images.

Fig. (12). Precision of the four methods applied on nucleus images.

Fig. (13). Recall of the four methods applied on nucleus images.

cell contours are blurred. The segmentation results of k-

means, Otsu’s, and GMAC (Fig. 14b, d, e) seem to be

washed out. Using the nucleus, light areas, we can identify

the existing cells in the image. The segmentation result of

GMM EM (Fig. 14c) is still under-segmented. In Fig. (15),

we can see that the background, denoted by the annotated

As shown in Fig. (14a), brain cell images are dark, and

0.4

0.5

0.6

0.7

0.8

0.9

1

1357911 1315

Image Number

F-Score

K-means

Otsu's

EM

GMAC

0

0.2

0.4

0.6

0.8

1

1.2

1 2 3 4 5 6 7 8 9 1011 1213 141516

Image Number

Precision

K-means

Otsu's

EM

GMAC

0

0.2

0.4

0.6

0.8

1

1.2

1357911 1315

Image Number

Recall

K-means

OTSU

EM

GMAC

Page 8

48 The Open Medical Informatics Journal, 2010, Volume 4 Du and Dua

line marked by ‘*’, has a narrow estimated intensity

distribution while cells distribute in a wider intensity levels,

denoted by the other annotated line marked by ‘+’. This

presentation using standard Gaussian distribution leads to

errors in the estimate of probability distribution. As shown in

Fig. (16), the individual intensity distributions are summed

to obtain the mixed Gaussian distribution, which is presented

by the annotated line marked by square. The errors in

estimation are accumulated in the sum procedure, which can

be presented by the discrepancy between the areas covered

by the estimated distribution and the true intensity

distribution denoted by the annotated line marked by

triangle. The other three segmentation results have cells split

with the nucleus, although several cells are over-segmented.

This over-segmentation can be explained as that these

techniques consider image intensity and texture information

in the segmentation process, while the spatial relation or

some connection between pixels is missed. Moreover,

compared to the synthetic images in Section 3.1.1, real cell

images are more complex and difficult to segment.

Fig. (15). Estimated intensity distribution of image in Fig. (14a)

using GMM EM model.

4. CONCLUSION

image segmentation. The four methods are compared and

contrasted to showcase efficacy strengths, as well as

embedded limitations. While no single method outperforms

the others in all tests, this analysis is expected to assist image

scientists in improving these techniques for the more

We present four unsupervised mining methods in cell

complex cell image segmentation problems encountered in

related disciplines.

Fig. (16). The ground truth and estimation of the intensity

distribution of image in Fig. (14a).

qualitatively using synthetic simulated and real images. EM

performs weakly in both cases due to its presumed Gaussian

model. It needs a better model assumption in microscopic

imaging if applied in cell image segmentation. Otsu’s

method cannot always guarantee a good segmentation result,

especially when the contrast between the background and

cells is poor. GMAC integrates intensity and gradient

information and keeps a stable performance in our experi-

ments. K-means can perform robust segmentation with the

aid of power function. In future work, spatial information

between pixels must be involved to improve the performance

of those techniques. The knowledge about the cell images,

such as inclusion of the power distribution function will be

incorporated in segmentation.

The methods are evaluated both quantitatively and

REFERENCES

[1]

Jean RP, Gray DS, Spector AA, Chen CS. Characterization of the

nuclear deformation caused by changes in endothelial cell shape. J

Biomed Eng 2004; 126(5): 552-8.

Osher S, Sethian JA. Fronts propagating with curvature-dependent

speed: Algorithms based on Hamilton-Jacobi formulations. J

Comput Phys 1988; 79: 12-49.

Ohlander R, Price K, Reddy DR. Picture segmentation using a

recursive region splitting method. Comput Graph Image Process

1978; 8: 313-33.

Jain AK. Data Clustering: 50 Years Beyond K-Means. Technical

Report TR-CSE-09-11. Pattern Recognit Lett 2009; in press.

[2]

[3]

[4]

a) b) c) d) e)

Fig. (14). a) Brain cell microscopy image, b) K-means clustering result, c) EM clustering result, d) Otsu’s segmentation result, e) GMAC

segmentation result.

050100

Intensity levels

150200250

0

0.02

0.04

0.06

0.08

0.1

0.12

Estimated probability of intensity distribution

Intensity distribution of background

Intensity distribution of cells

050100

Intensity levels

150200250

0

0.02

0.04

0.06

0.08

0.1

0.12

Intensity distribution

True image intesity distribution

Estimated image intensity distribution

Page 9

Segmentation of Fluorescence Microscopy Cell Images Using Unsupervised Mining The Open Medical Informatics Journal, 2010, Volume 4 49

[5]

Cootes T, Taylor CJ, Cooper DH, Graham J. Active shape models-

Their training and application. CVGIP: Image Understanding 1995;

61: 38-59.

Pham ZL, Xu C, Prince JL. Current methods in medical image

segmentation. Ann Rev Biomed Eng 2000; 2: 315-37.

Wahlby C, Lindblad J, Vondrus M, Bengtsson E, Bjorkesten L.

Algorithms for cytoplasm segmentation of fluorescence labelled

cells. Anal Cell Pathol 2002; 24(2-3): 101-11.

Jones TR, Carpenter A, Golland P. Voronoi-based segmentation of

cells on image manifolds. Lect Notes Comput Sci 2005; 535-43.

Bazi Y, Rruzzone L, Melgani F. Image thresholding based on the

EM algorithm and the generalized Gaussian distribution. Pattern

Recognit Lett 2007; 40: 619-34.

Otsu N. A threshold selection method from Gray-level Histogram.

IEEE Trans Syst Man Cybernetics 1979; 1; Vol. SMC-9.

Yan PK, Zhou XB, Shah M, Wong STC. Automatic segmentation

of high-throughput rnai flurescent cellular images. IEEE Trans Inf

Technol Biomed 2008; 12(1): 109-17.

Coulot L, Kischner H, Chebira A, et al. Topology preserving

STACS segmentation of protein subcellular location images. In:

Proc IEEE Int Symp Biomed Imaging, Arlington, VA, Apr 2006;

pp. 566-9.

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

Zimmer C, Labruyere E, Meas-Yedid V, Guillen N, Olivo Marin J-

C. Segmentation and tracking of migrating cells in Videomirco-

scopy with parametric active controus: a tool for cell-based drug

testing. IEEE Trans Image Process 2002; 12(10): 1212-21.

Benchmark set of synthetic images for validationg cell image

analysis algorithms: Benchmark images. Available from:

http://www.cs.tut.fi/sgn/csb/simcep/benchmark/

September 2009].

Cell Profiler: Cell image analysis software. Available from:

http://www.cellprofiler.org/ [Accessed: 10 September 2009].

Bresson X, Esedoglu S, Vandergheynst P, Thiran J, Osher S. Fast

Global Minimization of the Active Contour/Snake Model. J Math

Imaging Vis 2007; 28(2): 151-67.

Chan TF, Vese LA. Active contours without edges. IEEE Trans

Image Process 2001; 10(2): 266-77.

Ruusuvuori P, Lehmussola A, Selinummi J, Rajala T, Huttunen H,

Yli-Harja O. Benchmark set of synthetic images for validating cell

image analysis algorithms. In: Proceedings of the 16th European

Signal Processing Conference (EUSIPCO-2008), Lausanne,

Switzerland, 2008.

[14]

[Accessed: 10

[15]

[16]

[17]

[18]

Received: October 10, 2009

Revised: November 15, 2009 Accepted: November 15, 2009

© Du and Dua; Licensee Bentham Open.

This is an open access article licensed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-

nc/3.0/) which permits unrestricted, non-commercial use, distribution and reproduction in any medium, provided the work is properly cited.

Page 10

ABOUT BENTHAM

BENTHAM OPEN

EDITORIAL POLICIES

REVIEWERS

GUIDELINES

MEMBERSHIP

JOURNALS A - Z

JOURNALS BY

SUBJECT

INDEPENDENT

JOURNALS

ORDER PRINT VERSION

ADVERTISING DETAILS

SEARCH

Home

Aims & Scope

Editorial Board

Special issues

Manuscript Submission

& Instructions

Order Printed Reprints

Endorsements

View Published Contents

VIEW JOURNAL ARTICLES

ON PUBMED CENTRAL

Tue Nov 26 2013

Advertisement

The

Medical

Informatics

Journal

Open

Aims & Scope

The Open Medical Informatics Journal is an

Open Access online journal, which publishes

research articles, reviews, letters and guest edited

single topic in all areas of medical informatics.

The Open Medical Informatics Journal, a peer-

reviewed journal, is an important and reliable

source of current information on developments in

the field. The emphasis will be on publishing

quality papers rapidly and freely available to

researchers worldwide.

© BENTHAM OPEN

11/26/2013http://www.benthamscience.com/open/tominfoj/AimsScope.htm