Optik 121 (2010) 1752–1755 ﾧ

Contents lists available at ScienceDirect

Optik

journal homepage: www.elsevier.de/ijleo ﾧ

Study on the methods of image enhancement for liver CT images

a a, b
Li Yang , Yanmei Liang , Hailun Fan
a b
Key Laboratory of Opto-electronic Information Science and Technology, Ministry of Education, Institute of Modern Optics, Nankai University, Tianjin 300071, PR China Department of
Radiology, Tianjin Medical University General Hospital, Tianjin 300052, PR China
article info

Article history:
Received 16 December 2008
Accepted 21 April 2009

Keywords:
Liver CT image
Image segmentation
Image enhancement

1. Introduction

As a medical imaging technique, computed tomography (CT) is quite useful for doctors to analyze the pathological changes of the biological organs.
In order to reduce deaths, the diseases must be detected accurately in the early stage. However, the inherent low-density resolution makes CT images for
the biological parenchy-ma, such as the liver, often have low contrast, which sometimes makes doctors judge inaccurately. Some contrast agents can be
injected to enhance the CT image, but they are harmful to the patient, and lethal to some patients because of anaphylaxis.

Medical image processing, as an assistant means, can play an important role in early diagnosis. Some processing methods have been proposed to
segment the related organ from the whole medical images [1–3] or analyze and classify the texture [4–6]ﾧ . However, because of the complicated
textures and low gray differences in liver CT images, for some early local pathological changes, such as liver cancers, hepatic abscess, etc., the changes
of gray levels are too small to be noticed, it would be a great challenge even for some experienced doctors to make a right diagnosis.

Currently, there are some image enhancement methods. Histogram equalization (HE) is a common way to improve contrast, which generates an
image whose pixels of gray levels are as equal as possible. But for the original CT image, because of the large low gray background, HE often makes the
whole image too bright to see [7,8]ﾧ. In recent years, contrast limited adaptive histogram equalization (CLAHE) has been used in medical images

Corresponding author.
E-mail address: [email protected] (Y. Liang).
 In this paper, a two-step method is proposed to effectively enhance the liver region. The liver region is firstly segmented from the CT image with a
simple segmentation method, certain algorithm is then used to enhance the contrast of the segmented liver image. Several enhancement algorithms are
explored and compared. From the experimental results, the liver focus can be obviously distinguished from the normal liver tissue, which will help
doctors to diagnose the forepart focus correctly.

2. The methods of image enhancement

Generally, the gray levels of a digital CT image cover from 0 to 255, but the gray levels of the liver region cover a small range, so it is difficult to do
enhancement on the original digital CT image effectively. In this paper, the liver region is firstly segmented from the original image to form a new image
which only includes the liver region and a black background. Then the enhancement algorithms are done on the segmented image.

2.1. Segmentation of the liver CT image

Because the liver is the biggest organ in the liver CT image, a simple but effective way to segment the liver is used to reduce the processing time. The
segmentation procedures are as follows:

(1) Find out the gray level that appears most in the original image except the gray level 0 and 255.
(2) Take the gray level as a center, reserve the gray levels within a certain radius around the center level, and set other pixels’ gray level as 0. Then,
some discrete blocks in the CT image are
0030-4026/$ - see front matter Crown Copyright & 2009 Published
by Elsevier GmbH. All rights reserved.
doi:10.1016/j.ijleo.2009.04.006
L. Yang et al. / Optik 121 (2010) 1752–1755 1753
 normal school, the gray healthy part
range [a b, a+b] in the gray without careful
obtained. The above two steps can be called gray levels move to the low gray levels.
level can cover about 84% observation.
filtering. The formula of this
pixels of the liver, the range transforma-tion is described
[a 2b, a+2b] can cover about as follows:
(3) Reserve the block whose area is the biggest and set the
99% pixels of the liver, and The curve of
others as 0.
[a 3b, a+3b] will cover about 0
I ¼ C log10ð1 þ jI the gray levels
all of the liver. from an
(4) Set the reserved block’s gray levels as 255. Execute
arbitrarily
morpho-logic close operation and fill the holes in the IminjÞ; ð3Þ selected row
block using gray level 255. The liver denoted with the
(3) Nonlinear stretching with marked by a
gray level 255 is segmented. 0 horizontal line in
the logarithmic where I and I are the gray
(5) Fill in the regions which have gray level 255 using the
transformation: If the levels before and after Fig. 1 ﾧ (d) is
0 transforma-tion,
gray levels in original CT image in the same region. The transform function I ¼ T(I) shown in Fig.
respectively. C is a constant
is a nonlinear single-value
nearly entire liver region denoted with its original gray which determines the range 2 ﾧ. It can be
func-tion, the transformation of the gray levels after
levels is obtained from the CT image.
is called gray nonlinear transformation; Imin is the
transforma-tion. The minimum gray level of the
logarithm transformation is liver region before
often used to stretch low transformation.
2.2. Enhancement algorithms gray level, and compress (4) Selective histogram
high gray level. Therefore, equalization: The
the details of low gray conventional histogram
Several enhancement algorithms are adopted in this paper,
region can be distinguished equalization (HE) is to
which are described as follows:
more clearly. Here the
enhance the whole image,
minimum gray level in the
(1) Direct stretching with the linear relationship: The so if there are large low
liver region will be
purpose of the image enhancement is to improve the subtracted from the image gray level regions in the
image contrast, which transforms the gray range in the before transformed and thus image, the enhanced result
original image into a all gray levels will will be too bright and
0 cannot improve the visual
relatively larger range. If the transform function I ¼ T(I)
effect. The method of
is a linear single-valued function, this treatment is called
selective histogram
gray linear transform, or linear stretching, which can be
performed with the following formula by direct equalization is to choose the
stretching: interested regions to do
histogram equalization and
0 0 0 0
I I I I I I leave others alone.
maxmin

Þþ ;
¼ Imax Imin ð min min

Based on the segmented liver

image, the histogram of the liver
0
region can be obtained easily.
where I and I are the gray levels before and after Then executing histogram
0 0 Fig. 1. A liver CT
transforma-tion, respectively. I max and I min are the equal-ization on this histogram image. (a) An
highest and lowest gray level after transformation, Imax enhances the liver region. original liver CT
image; (b) the result
and Imin are the maximum and minimum gray level in the after gray level
liver region before the transformation, respectively. filtering; (c) the
3. Experimental results and result after the close
(2) Linear stretching according to the fitting curve: The discussions operation and filling
histogram of the segmented liver CT region is more like the holes; (d) the
final segmented
a normal distrubution [9]ﾧ , so Gaussian function can be A 512 512 liver cancer CT
liver image (the
used to fit the histogram in order to determine the range image is used in this paper. The region in and around
which should be transformed. The Gaussian function is original image is shown in Fig. the real rectangular
box represents the
described as
1 ﾧ (a). The image after gray focus).

PðxÞ ¼ ce ðx 2
aÞ =b
2
: levels filtering is shown in Fig.

ð2Þ 1 ﾧ (b), where the center gray

level is 161, and the reserved
The parameter a represents the position of the peak radius of gray level is 30. The
center, b controls the width of the curve, and c is the image after the close operation
height of the curve’s peak. In this paper, the algorithm of and the segmented liver image
the least squares fitting is used to fit the histogram of the are respectively shown in Figs.
segmented liver CT region [10] and obtain the values of 1 ﾧ (c) and 1(d). From Fig. 1
a, b, and c. Then Eq. (1) is used to stretch the gray range.
ﾧ (d), the gray difference
The gray levels of the image lower than the minimum of
between the focus marked by a
the transformed range will be set as 0, and the levels
real line rectangular box and the
higher than the maximum of the transformed range will
healthy part is inconspicuous, so
be set as 255. The transformed range [Imin, Imax] is the cancer region cannot be
depended on a and b. According to the definition of the clearly distinguished from the
1754 L. Yang et al. / Optik 121 (2010) 1752–1755
 of which a dashed minimum and
the maximum of
rectangular box the gray levels
represents the healthy of the liver. The
minimum and
part and the real the maximum
rectangular box are easy to be
influenced by
represents the focus. the noise. So the
The average gray range may be
levels of the two too large or too
small because of
regions, Ihealthy and the noise. A wide
Ifocus, are calculated. initial stretching
range may result
The in an incon-
spicuous
contrast between them is given by the formula improvement,
Fig. 2. The curve of the gray levels from an arbitrarily selected row shown while a narrow
by a white line inFig. 1 ﾧ (d). The focus is signed with a thin curve and
Contrast ¼
I
healthy
I
focus
:
stretching range
the healthy part is signed with the thick curves. may result in
I I

þ over
healthy focus
enhancement
and lose the
microstructure
The same regions as Fig. 1 of the focus
texture.
ﾧ (d) in Figs. 4 ﾧ (a)–(f) are
used to calculate their contrasts.
The results are shown in Table
1 ﾧ. The
linear
stretching
with
Gaussian
fitting in
the range
of [a b,
Fig. 3. The distribution of gray levels in the liver region with dots and its
fitting Gaussian curve with a blue line. The Y-axis presents the normalized Fig. 4. The liver CT enhanced images. a+b] has
probability density of the gray levels. (a) Direct stretching with the linear
relationship, (b) logarithmic
the largest
transformation (C ¼ 3.075), (c) linear contrast
stretching with Gaussian fitting in the
range of [a b, a+b], (d) linear stretching enhanceme
seen that the gray levels of the focus are slightly less than
with Gaussian fitting in the range of [a
2b, a+2b], (e) linear stretching with
nt, but the
those of the healthy part around it. The distribution of gray Gaussian fitting in the range of [a 3b, stretching
a+3b], (f) the selective histogram
levels of the liver region of Fig. 1 ﾧ (d) is given in Fig. 3 equalization.
with dots, and the fitting result with Gaussian function is
expressed with a blue curve.
Fig. 1 ﾧ (d) is enhanced by the methods Table 1
Contrast of the original image and the
described in Section 2.2. The results are shown enhanced images.
in Figs. 4 ﾧ(a–f). Where C is chosen 3.075 in Methods Fig. 1(d) Fig. 4(a) Fig. 4(b)
logarithmic transformation; a is 162.7; and b is
16.01 in Gaussian fitting from Fig. 3 ﾧ . The Contrast 0.04 0.11 0.10

highest and lowest gray level after

transformation, I0max and I0min, is 255 and 0
for all these methods, respectively. From Table 1 ﾧ ,
compared with the original
The quality of the liver image has image, the processed images
have been enhanced in different
been improved from Fig. 4 ﾧ . The degrees. The linear and
contrast between the healthy part and the logarithm stretching are easy to
execute. The initial stretching
focus is used to evaluate these methods. range in linear and logarithm
Two regions are selected in Fig. 1 ﾧ (d), stretching is just selected by the
L. Yang et al. / Optik 121 (2010) 1752–1755 1755
 [8] J.A. Stark, IEEE Trans. Image
Process. 9 (2000) 889–896.
the enhanced result shown in [9] S.J. Lim, Y.Y. Jeong, Y.S. Ho,
Advances in Multimedia
Fig. 5 ﾧ (b), the peripheral Information Processing-PCM
walls of the abscess lesions 2005, vol. 3767, Springer, Berlin,
Heidelberg, 2005.
enclosed by the rectangular box [10] Y. Liu, Y. Liang, Z. Tong, X. Zhu,
are clearer than the unprocessed G. Mu, Opt. Commun. 279 (2007)
image. 23–26.

4. Conclusion

The methods of image

Fig. 5. A hepatic abscess CT image. (a) The original CT image, (b) the result enhancement based on the
of linear stretching with Gaussian fitting in the range of [a 3b, a+3b] (a ¼ segmented liver CT images are
156.60, b ¼ 15.76). performed, which can effectively
stretch the gray of liver region
and highlight the lesions. In all
the transformations, the selective
range is appreciably small, and introduces a little over histogram equalization, linear,
enhance-ment. The linear stretching with Gaussian fitting in and logarithm stretching are
the range of [a 2b, a+2b] and [a 3b, a+3b] have smaller simple and easy to execute, but
contrast enhance-ment than that of the range of [a b, a+b], but the range of transformation can
they have better performances, which the microstructure of be obtained adaptively by fitting
the focus texture can be clearly distinguished. The range the histogram with the Gaussian
which is determined by Gaussian fitting is based on the whole curve, which can achieve a better
gray levels of the liver region, so it can reduce the influence effect according to different CT
of the noise in a certain extent. images. Distinguishing the focus
automatically with some
algorithms based on the
The selective histogram equalization has the second enhanced images will be our
largest contrast enhancement, and this algorithm is also based future work.
on the whole gray levels of the liver region, so it can reduce
the influence of the noise.

In practice, for selecting the stretching range, there is a

trade-off between reducing the noise in the image and Acknowledgements
avoiding over enhancement. In addition, the gray levels of the
CT image are proportional to the densities of the organ, so all This research is supported
enhanced methods should only change the distances between the National Natural Science
gray levels, but cannot change their orders in the histogram. Foundation of China (Grant No.
60677012) and the Tianjin
Another 512 512 hepatic abscess CT image is shown in Foundation of Natural Science
Fig. ﾧ 5 ﾧ (a). It is segmented, by using the (Grant No. 09JCZDJC18300).

procedure of Section 2.1, of which the center

References
gray level is 158, and the reserved radius of
gray level is 30. The segmented liver CT image [1] F. Liu, B. Zhao, P.K. Kijewski, L.
is enhanced by linear stretching with Gaussian Wang, L.H. Schwartz, Med. Phys.
32 (2005) 3699–3706.
fitting in the range of [a 3b, a+3b]. From [2] A. Yezzi, S. Kichenassamy, A.
Kumar, P. Olver, A. Tannenbaum,
IEEE Trans. Med. Imaging 16
(1997) 199–209.
[3] S. Wang, D. Fu, M. Xu, D. Hu,
Artif. Intell. Med. 39 (2007) 65–
77.
[4] A.H. Mir, M. Hanmandlu, S.N.
Tandon, Eng. Med. Biol. Mag.
IEEE 14 (1995) 781–786.

[5] S.M. Yamany, K.J. Khiani, A.A.

Farag, Pattern Recognition Lett. 18
(1997) 1205– 1210.

[6] S. Mougiakakou, I. Valavanis, A.

Nikita, K. Nikita, Artif. Intell.
Med. 41 (2007) 25–37.

[7] Y. Jin, L. Fayad, A. Laine, SPIE

Proc. 4478 (2001) 206–213.