Liver segmentation using marker controlled watershed transform

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 13, No. 2, April 2023, pp. 1541~1549
ISSN: 2088-8708, DOI: 10.11591/ijece.v13i2.pp1541-1549  1541
Journal homepage: http://ijece.iaescore.com
Liver segmentation using marker controlled watershed
transform
Kiran Malhari Napte, Anurag Mahajan
Department of Electronics and Telecommunication, Symbiosis Institute of Technology, Symbiosis International (Deemed University),
SIU, Lavale Pune, India
Article Info ABSTRACT
Article history:
Received Mar 30, 2022
Revised Sep 20, 2022
Accepted Oct 15, 2022
The largest organ in the body is the liver and primarily helps in metabolism
and detoxification. Liver segmentation is a crucial step in liver cancer
detection in computer vision-based biomedical image analysis. Liver
segmentation is a critical task and results in under-segmentation and
over-segmentation due to the complex structure of abdominal computed
tomography (CT) images, noise, and textural variations over the image. This
paper presents liver segmentation in abdominal CT images using
marker-based watershed transforms. In the pre-processing stage, a modified
double stage gaussian filter (MDSGF) is used to enhance the contrast, and
preserve the edge and texture information of liver CT images. Further,
marker controlled watershed transform is utilized for the segmentation of
liver images from the abdominal CT images. Liver segmentation using
suggested MDSGF and marker-based watershed transform help to diminish
the under-segmentation and over-segmentation of the liver object. The
performance of the proposed system is evaluated on the LiTS dataset based
on Dice score (DS), relative volume difference (RVD), volumetric
overlapping error (VOE), and Jaccard index (JI). The proposed method gives
(Dice score of 0.959, RVD of 0.09, VOE of 0.089, and JI of 0.921).
Keywords:
Computer tomography
Gaussian filtering
Image enhancement
Liver segmentation
Watershed transform
This is an open access article under the CC BY-SA license.
Corresponding Author:
Anurag Mahajan
Department of Electronics and Telecommunication, Symbiosis Institute of Technology, Symbiosis
International (Deemed University)
Pune, Maharashtra, India
Email: anurag.mahajan@sitpune.edu.in
1. INTRODUCTION
The liver is the third and seventh cause of causality due to cancers across the globe. In 2020 it is
estimated that 30,230 and 9,930 males and females are died due to liver cancer. The liver cancer rate is
steadily increasing by 2% since 1970 [1]. The largest organ in the human body, the liver, is situated on the
right side of the abdomen. The liver's primary job is to filter blood coming from the digestive system before
it is sent to the rest of the body. It also helps in the metabolism and detoxification of chemicals and drugs.
The liver also produces proteins that are important for blood clotting [2].
Primary and secondary liver cancers are the two subtypes of liver cancer. Primary liver cancer is
brought on by deviations in the various liver cells [3]. Primary liver cancers are categorized as hepatocellular
carcinoma (HCC), intrahepatic cholangiocarcinoma, hemangiosarcoma, angiosarcoma, and hepato-blastoma
[4]. HCC is most general and accounts for 75% of all liver cancers. Secondary liver cancer originates due to
the cells from other body parts such as the stomach, lungs, breast, colon, and pancreas. Secondary liver
cancers are also called metastatic liver cancer [5]. Liver cancer may cause due to heavy alcohol use, smoking,
family history, obesity diabetes, hepatitis B or C virus, and low immunity. Some of the common symptoms

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 13, No. 2, April 2023: 1541-1549
1542
of liver cancers are abdominal pain, weight loss, jaundice, abdomen swelling, nausea, fatigue, back pain,
vomiting, fever, or itching [6]. Figure 1 shows the liver anatomy in Figure 1(a), normal liver computed
tomography (CT) image in Figure 1(b), and liver cancer CT images in Figure 1(c).
(a) (b) (c)
Figure 1. The liver anatomy: (a) liver structure, (b) normal liver CT image, and (c) liver cancer CT image
Liver cancer is more common in males compared to females and in people over age so. Liver cancer
can be diagnosed by measuring bilirubin, enzymes, and proteins alpha-protein (AFP) in blood, and through
liver biopsy. In the early years, liver cancer analysis used to be performed using various manual tests such as
cytotoxic chemotherapy, immunotherapy, or oncolytic virus therapy [7]. Liver segmentation plays vital role
in the liver tumor detection in the computer vision However, due to the liver's uneven and excessive
segmentation in the abdomen CT scans, tumor segmentation is difficult. Many times, because of inadequate
segmentation the borderline liver tumor detection is challenging. Thus, there is need to focus on the liver
segmentation to improve the liver cancer detection.
Computer vision-based biomedical image processing has attracted the attention of researchers
toward liver cancer segmentation and detection using various images such as ultrasonic images magnetic
resonance imaging (MRI), single-photon emission computed tomography (SPECT), and positron emission
tomography (PET). CT scan is a faster method of capturing biomedical images and it can give images of
tissues, skeletal structure, and organs. Thus, CT images are widely used for the analysis of abdominal parts of
the body. Liver segmentation in the abdominal CT images is a critical task because of a low contract between
variable liver areas and the neighboring organs such as the stomach, kidney, pancreas, and muscles [8], [9].
Abdominal CT images consist of various body parts along with the liver. To extract the correct properties of
the liver, the liver portion of the CT images must be segmented. In the past, a variety of ways for separating
the liver from abdominal CT scans have been reported. These techniques can be divided into three categories:
intensity-based, structure-based, and machine learning or deep learning techniques.
This paper presents, effective liver segmentation using a modified double stage gaussian filter
(MDSGF) for liver image enhancement and a marker-controlled watershed algorithm for liver segmentation.
The objectives of the paper are summarized as follows: i) Implementation of proposed MDSGF for liver
image enhancement to tackle the problem of noise, and complex structure of the abdominal region;
ii) Implementation of marker controlled watershed algorithm for liver segmentation to analyze the effect of
proposed enhancement on the segmentation; and iii) Performance evaluation of proposed scheme on the LiTS
dataset based on dice score (DS), volume overlapping error, relative volume difference, Jaccard index.
The remaining paper is structured as follow: section 2 focuses on the related work on liver
segmentation based on various computer vision based techniques. Section 3 gives a brief discussion of the
proposed methodology that encompasses the pre-processing and segmentation of liver CT image. Section 4
provides the experimental results and discussion of liver cancer detection. Finally, section 5 concludes the
paper and gives the future direction for the improvement of the work.
2. LITERATURE REVIEW
Various computer vision based methods have been presented for the liver segmentation in last two
decades. The intensity-based techniques are the active contour model [10], region growing technique [11],
super-pixel technique [12], graph cut method [13], clustering-based techniques [14], and machine learning-
based techniques. Intensity-based techniques consider the gray intensity value of the pixel with the

Int J Elec & Comp Eng ISSN: 2088-8708 
Liver segmentation using marker controlled watershed transform (Kiran Malhari Napte)
1543
neighborhood pixel intensity to decide the texture of the liver. In the techniques, the seed points are manually
placed by the expert in the liver parenchyma. The pixel that matches with the seed pixels is clustered
collectively to form the homogeneous texture region. Because the shape control capability was not available,
intensity-based approaches frequently produced leakage in the manually seeded area, rough ridges, and
edges. Intensity-based approaches have been found to be ineffective for magnetic resonance pictures with
very diverse textures [15]. In the graph-cut technique, the liver part is roughly considered as the foreground
and the peri-hepatic structure is considered as the background part. Then using cut methods the foreground
and background pixels are separated into the homogeneous regions [16]. Machine learning-based methods
such as support vector machines (SVMs) and artificial neural networks (ANN) [15] have been successfully
presented for the quantitative feature extraction of the image texture. Recently, deep learning has attracted
the wide attention of researchers for medical image segmentation because of its higher discriminative power,
representation capability, and internal connectivity. Fan et al. [17] presented multi-scale context nested UNet
(MSN-Net) for the liver segmentation to diminish the gradient vanishing problem and lessen the semantic
gap. It consists of a multi-scale context fusion network that combined high and low-level features to improve
increases the complexity of the network. Bai et al. [18] presented a multi-scale candidate generation method
(MCG), 3D fractal residual network (3D FRN), and an active contour model (ACM) to deal with varying size,
shape, and location of the tumor. It suffered from accurate detection of multiple tumor regions separately and
borderline tumors. Li et al. [19] investigated hybrid densely connected UNet (H-DenseUNet) for accurate and
automatic HCC detection. The use of parallel training makes the system extremely complex. Wang et al. [20]
investigated the hybrid dilated and attention residual-UNet (HDA-ResUNet) for the contextual liver
segmentation. It showed DS, of 0.94 but suffers from poor balancing of computational resources and spatial
dimensions. DL-based algorithms have given better performance and highly discriminative capability than the
intensity-based methods. The performance of DL-based techniques is highly dependent on noise, contrast,
shape, structure, borders, and the type of imaging technique for the liver image.
Watershed transform is a region-based segmentation technique based on mathematical morphology
and geographical information of the image. Watershed transform has attracted wide attention towards
medical image segmentation because of its simplicity, lesser computational steps, and ability to segment
smaller as well as larger objects in the image. The traditional watershed algorithm often results in over-
segmentation of the objects in medical images. To minimize the over-segmentation of the liver in medical
magnetic resonance (MR) images, Lawankar et al. [21] investigated a marker watershed transform algorithm
for the liver segmentation along with a Weiner filter for noise removal in images. It has given poor
performance because of the spatial invariance of the Weiner filter. Abdallah [22] has shown that marker-
controlled watershed transformation can work better in presence of speckle noise and can eliminate the
problem caused due to contrast and boundaries of the liver. Huynh et al. [23] proposed a watershed algorithm
for liver segmentation and an active contour model for refinement of the segmentation. Anter et al. [24]
presented an adaptive watershed transform for the liver tumor segmentation that tackles the problem of high
noise, large variance, and fuzziness due to the same intensity regions of the tumor. They have used
neutrosophic sets (NS), and fast fuzzy c-mean clustering algorithms (FFCM) along with watershed transform
to improve the segmentation results. They have used a smaller dataset for the experimentation whereas the
results for the larger dataset were not explored. Watershed-based algorithms have given the better
performance and highly discriminative capability than other segmentation techniques. Very few researchers
have given concentration on the enhancement and filtering of liver CT images which limits the performance
of the liver segmentation in noisy and high contrast images. Sometimes, these techniques may result in
leakage or coarse segmentation due of the noise and rotation. Still, there is scope for improvement in the
watershed algorithm to avoid over-segmentation and to reduce the noise effect [25]–[27].
3. PROPOSED METHOD
The flow diagram of the proposed marker-based watershed transform for the liver segmentation is
shown in Figure 2. The proposed system is evaluated on the LiTS dataset which consists of 200 CT volumes
with different contrast levels, tissue abnormalities, and liver size [28]. The 2-D images are selected from the
specific volume of original CT data having complete view of the all-abdominal organs to minimize the
computational efforts.
The pre-processing stage includes the separation of the liver image into three channels such as red,
green, and blue. Previous liver segmentation has given very little concentration on the liver image filtering
which degrades the performance of the liver segmentation. Nidhyanandhan et al. [29] proposed the DSGF, a
double stage Gaussian filter for improving underwater images. This method needs visual correction after
principal component analysis (PCA) fusion which degrades the edge information of the liver CT images. In
this paper, we propose a MDSGF that considers Wavelet transform-based fusion for the liver image
enhancement to preserve the edges of the liver CT images as illustrated in Figure 3.

 ISSN: 2088-8708
1544
Figure 2. Flow diagram of proposed liver segmentation technique
Figure 3. Flow diagram of proposed MDSGF
In the proposed MDSGF, the original liver CT image is separated into red, green, and blue color
channels. In the first stage, a Gaussian filter is applied two times for the individual color channel. Gaussian
filter is a low pass filter that eliminates the high-frequency components present in the lever CT image. The
Gaussian filtering for the two-dimensional liver CT image is shown in (1) [30].
𝐺(𝑥, 𝑦) =
1
2𝜋𝜎2 𝑒
−
𝑥2+𝑦2
2𝜎2
(1)
where, σ stands for the standard deviation of the Gaussian distribution, 𝑥 and 𝑦 represent the position of
pixels in the Gaussian window. The Gaussian filter with different window sizes is convolved with the
original image to get the smooth borders in the stomach CT image [31]. Three different values of variance
(σ) are selected for each channel Gaussian filter. Higher (0.9), low (0.3), and medium (0.5) are selected for
the Gaussian filtering of red, green, and blue color channels respectively. In the second phase, contrast
limited adaptive histogram is applied to each color channel to enhance the contrast of each color channel.
Further, median filtering is applied to smooth the texture of the liver image and minimize the non-uniform
components in the liver image. The two images are fused using the Wavelet transform where these images
are decomposed into four components such as approximation, diagonal, horizontal, and vertical components.
We have used the Haar wavelet filter for the wavelet decomposition. These four components of the two
images are combined using the mean fusion method. Then inverse wavelet transform is applied to reconstruct
the images to construct the enhanced image. The wavelet transform-based image fusion maintains the quality
of edges and textural information of the liver CT image [32].

1545
Using Otsu's thresholding, the grey image is transformed into a binary image. In this method, the
image histogram is split into two parts and the threshold is selected where the weighted variance of the
regions is minimum [33], [34]. The threshold value is selected using (2).
𝜎𝑤
2(𝑡) = 𝑤1(𝑡)𝜎1
2
+ 𝑤2(𝑡)𝜎2
2
(2)
where, 𝜎1
2
and 𝜎2
2
are the variance of the two regions. 𝑤1(𝑡) and 𝑤2(𝑡) are the probabilities of the two classes
divided by t which lies between 0 to 255. The binary image obtained using Otsu’s thresholding considers the
within-class variation over the lower intensity and high-intensity regions. The morphological opening is
performed to estimate the background region of the image [35].
The distance transform gives the distance of the pixel from the nearest border. This distance is used
to find the peak point and valley points in the image. The distance transform gives the connected components
in the image [36], [37].
Any gray-scale image is viewed as a tomographic surface that consists of peak points representing
the high-intensity pixels and the valley points representing low-intensity values [38]. Watershed transform is
based on filling the water in the isolated valley points which are represented by different label values. When
the water level is increased various valley points start to merge. To avoid the merging of these valley regions,
the regions are separated by border [39]. The process of water filling and border construction is continued till
all the peaks are covered. Watershed segmentation results in over-segmentation due to texture irregularity
and noise in the liver image. The marker-based watershed algorithm works on the estimation of the
foreground and background region of the image. The watershed transform is applied to the distance
transformed image. The marker image consists of the labels for different regions of the liver image. The
background region is labeled using 0, the foreground region is labeled with 1, and the border pixels which
separate different watershed regions are represented by -1 [40], [41].
4. RESULTS AND DISCUSSION
The proposed system is implemented using Python 3.8 and OpenCV toolbox on the personal
computer having a core i3 processor with 2.64 GHz speed, 4 GB RAM, and windows operating system. The
visualizations of the various stages of the proposed systems are shown in Figure 4. The outcomes of the proposed
liver enhancement and segmentation approach are evaluated on the LiTS dataset that encompasses 200 liver CT
images along with its ground truth images. All the images are having dimensions of 512×512 pixels.
Figure 4(a) shows original liver CT image sample from LiTS dataset, Figure 4(b) provides the
filtered image using proposed MDSGF, Figure 4(c) shows the binary image output obtained using Otsu’s
thresholding and Figure 4(d) shows morphologically filtered image with minimize the noise in binary image.
The distance transform images shown in Figure 4(e) is used to find peak and valley points to construct
watershed image as given in Figure 4(f). Figure 4(g) shows the watershed ridge lines, Figures 4(h) and 4(i)
shows the segmented liver region and liver area respectively. It helps to minimize the under and over
segmentation of the liver object.
Figure 4 shows that the proposed MDSGF along with marker controlled watershed provides better
segmentation of the boundary region of the liver area from the abdominal CT image. It helps to minimize the
under and over segmentation of the liver object. The performance of the proposed MDSGF is compared with
a simple Gaussian filter and DSGF based on mean square error (MSE), peak signal to noise ratio (PSNR),
and structural similarity index measure (SSIM) as provided in (3) to (5). It is found that the proposed
MDSGF shows better PSNR, improved SSIM, and lower MSE as given in Table 1.
𝑆𝑆𝐼𝑀(𝑥, 𝑦) =
(2𝜇𝑥𝜇𝑦+𝑐1)(2𝜎𝑥𝑦+𝑐2)
(𝜇𝑥
2+𝜇𝑦
2+𝑐1)(𝜎𝑥
2+𝜎𝑦
2+𝑐2)
(3)
where, 𝜇𝑥 , 𝜇𝑦 stands for an average of the original and enhanced liver image, 𝜇𝑥
2
, 𝜇𝑦
2
stands for the variance
of the original and enhanced liver image, 𝜎𝑥𝑦 is the covariance of the original and enhanced image,
𝑐1 = (𝑘1𝐿)2
and 𝑐2 = (𝑘2𝐿)2
weak denominator stabilizer, 𝐿 is the dynamic range of liver image pixels (up
to 255 for an 8-bit image), 𝑘1 = 0.01 and 𝑘2 = 0.03. SSIM value ranges between 0 and 1. A larger value of
SSIM shows better enhancement of the image. SSIM is calculated over the window size (x, y) of 5×5 pixels.
𝑀𝑆𝐸 =
1
𝑚𝑛
∑ ∑ [𝐼(𝑖, 𝑗) − 𝑁(𝑖, 𝑗)]2
𝑛−1
𝑗=0
𝑚−1
𝑖=0 (4)
𝑃𝑆𝑁𝑅 = 10. 𝑙𝑜𝑔10 (
2𝐵−1
𝑀𝑆𝐸
) (5)

 ISSN: 2088-8708
1546
where, 𝐼(𝑖, 𝑗) 𝑎𝑛𝑑 𝑁(𝑖, 𝑗) are original and enhanced images, m and n stand for rows and columns of images,
and B is the number of bits per sample.
Table 1 shows that the proposed system gives SSIM of 0.89, PSNR of 50.34 dB, and MSE of 0.6 for
a 3×3 window which is superior to a simple Gaussian filter and DSGF. Increasing the window size
diminishes the fine structures in the liver image. The smaller local window (3×3) helps to capture the local
variations to distinguish the different objects in the CT image. The higher SSIM shows that the proposed
filtering scheme retains the structural content of the CT image that avoids the under or over segmentation of
the liver. It has given better performance in terms of SSIM, PSNR and MSE compared with single stage
Gaussian filter and DSGF for CT image enhancement.
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
Figure 4. Visualization of experimental results; (a) original image, (b) MDSGF filtered image, (c) Otsu’s
threshold image, (d) morphological filtered image, (e) distance transform, (f) watershed regions,
(g) watershed ridge lines, (h) segmented liver region, and (i) segmented liver area
Table 1. Performance of MDSGF (average value for LiTS dataset)
Method 3×3 Window
SSIM PSNR (dB) MSE
Gaussian Filter 0.85 49.08 0.8
DSGF 0.87 49.67 0.7
Proposed MDSGF 0.89 50.34 0.6

1547
The performance of the segmentation algorithm is evaluated based on segmentation performance
metrics such as DS, volumetric overlapping error (VOE), relative volume difference (RVD), and Jaccard
index (JI). The DS, VOE, RVD, and JI are computed using (6) to (9).
DS(A, B) =
2|A∩B|
|A|+|B|
(6)
𝑉𝑂𝐸 (𝐴, 𝐵) = 1 −
|𝐴∩𝐵|
|A U B|
(7)
𝑅𝑉𝐷 (𝐴, 𝐵) =
|𝐴|−|𝐵|
|B|
(8)
𝐽𝐼 (𝐴, 𝐵) =
𝐷𝑆
2−𝐷𝑆
(9)
where A and B represent the binary mask of the segmented liver and the ground truth image. A smaller value
of VOE represents better liver segmentation whereas a higher DS represents better segmentation results.
RVD is a measure of segmentation where a positive RVD value refers to over-segmentation whereas a
negative RVD value refers to under-segmentation the similarity between the segmented image and the
ground truth is measured by JI i.e. Jacard index.
The results of the proposed technique are compared with previous implementations of liver
segmentation such as watershed-active contouring [23] and adaptive watershed transform [24]. The
performance of the proposed method is also compared with recent deep learning-based liver segmentation
techniques such as MSN-Net [17], ACM-ResNet [18], H-DenseUNet [19] and HAD-ResUNet [20]. It is
observed that proposed MDSGF helps to increase the liver segmentation by smoothing the irregularities
present in the abdominal CT images as shown in Table 2. The proposed technique has shown better
improvement in DS and JI respectively over the previous segmentation techniques such as watershed-Active
Contouring, adaptive watershed transforms, ACM-ResNet, MSN-Net, HAD-ResUNet and H-DenseUNet. It
shows a slightly declined VOE compared with the watershed-active contouring technique and MSN-Net. The
use of MDSGF helps to preserve the edge, shape, and texture properties of the liver region help to segment
the liver area accurately using marker-controlled watershed transform, and gives a superior improvement in
DS, VOE, JI, and RVD respectively over the segmentation without MDSGF.
Table 2. Performance comparison of proposed method
Method DS VOE JI RVD
ACM-ResNet [18] 0.67 0.324 - 0.194
H-DenseUNet [19] 0.937 0.116 - -0.010
MSN-Net [17] 0.942 0.04 0.89 -
HAD-ResUNet [20] 0.940 - 0.89 -
Watershed+ Active contouring [23] 0.911 0.083 - -
Adaptive watershed transform [24] 0.928 - 0.868 -
Proposed method (With MDSGF) 0.959 0.089 0.921 0.09
5. CONCLUSION
Thus, this paper presents automatic computer vision-based liver segmentation using marker-
controlled watershed transform. The proposed marker-controlled watershed transform along with the
proposed novel modified double stage Gaussian filter tackles the problem of over-segmentation in
generalized watershed-based segmentation caused due to noise, complexity in liver structure, and contrast of
the liver image. The MDSGF of the CT images has shown improvement in the liver segmentation. The
proposed system has shown better performance for liver segmentation in noisy and complex structured
abdominal CT images. The Gaussian filter has shown improvement in the liver segmentation performance
compared to baseline algorithms. It has demonstrated resilience to both under- and over-segmentation of the
liver area on abdominal CT imaging. The proposed method results in a DS of 0.959, VOE of 0.089, JI of
0.921, and RVD of 0.09. The future scope of the proposed work consists of performance improvement of the
watershed segmentation technique using deep learning algorithms. Further, the proposed method can be
extended for liver tumor detection.
ACKNOWLEDGEMENTS
I would like to thank the Principal, Head of Department (E&TC) of Pimpri Chinchwad College of
Engineering and Research, Ravet for their continuous support and motivation for the research.

 ISSN: 2088-8708
1548
REFERENCES
[1] ASCO, “Liver cancer: statistics,” cancer.net, 2021. https://www.cancer.net/cancer-types/liver-cancer/statistics (accessed May 15,
2021).
[2] I. M. Arias et al., The liver: biology and pathobiology. John Wiley & Sons, 2020.
[3] M. Torbenson and P. Schirmacher, “Liver cancer biopsy - back to the future?!,” Hepatology, vol. 61, no. 2, pp. 431–433, Feb.
2015, doi: 10.1002/hep.27545.
[4] Healthline, “Liver cancer.” healthline.com, https://www.healthline.com/health/liver-cancer (accessed Apr. 10, 2021).
[5] D. Sia, A. Villanueva, S. L. Friedman, and J. M. Llovet, “Liver cancer cell of origin, molecular class, and effects on patient
prognosis,” Gastroenterology, vol. 152, no. 4, pp. 745–761, Mar. 2017, doi: 10.1053/j.gastro.2016.11.048.
[6] Mayo clinic, “Liver cancer.” mayoclinic.org, https://www.mayoclinic.org/diseases-conditions/liver-cancer/symptoms-causes
(accessed Apr. 10, 2021).
[7] C.-Y. Liu, K.-F. Chen, and P.-J. Chen, “Treatment of liver cancer,” Cold Spring Harbor perspectives in medicine, vol. 5, no. 9,
2015.
[8] F. A. Mohammed and S. Viriri, “Liver segmentation: A survey of the state-of-the-art,” in 2017 Sudan Conference on Computer
Science and Information Technology (SCCSIT), Nov. 2017, pp. 1–6, doi: 10.1109/SCCSIT.2017.8293049.
[9] A. Das, U. R. Acharya, S. S. Panda, and S. Sabut, “Deep learning based liver cancer detection using watershed transform and
Gaussian mixture model techniques,” Cognitive Systems Research, vol. 54, pp. 165–175, May 2019, doi:
10.1016/j.cogsys.2018.12.009.
[10] A. Hoogi, A. Subramaniam, R. Veerapaneni, and D. L. Rubin, “Adaptive estimation of active contour parameters using
convolutional neural networks and texture analysis,” IEEE Transactions on Medical Imaging, vol. 36, no. 3, pp. 781–791, Mar.
2017, doi: 10.1109/TMI.2016.2628084.
[11] O. F. Abd-Elaziz, M. S. Sayed, and M. I. Abdullah, “Liver tumors segmentation from abdominal CT images using region growing
and morphological processing,” in 2014 International Conference on Engineering and Technology (ICET), Apr. 2014, pp. 1–6,
doi: 10.1109/ICEngTechnol.2014.7016813.
[12] T. Kitrungrotsakul, X.-H. Han, and Y.-W. Chen, “Liver segmentation using superpixel-based graph cuts and restricted regions of
shape constrains,” in 2015 IEEE International Conference on Image Processing (ICIP), Sep. 2015, pp. 3368–3371, doi:
10.1109/ICIP.2015.7351428.
[13] G. Li, X. Chen, F. Shi, W. Zhu, J. Tian, and D. Xiang, “Automatic liver segmentation based on shape constraints and deformable
graph cut in CT images,” IEEE Transactions on Image Processing, vol. 24, no. 12, pp. 5315–5329, Dec. 2015, doi:
10.1109/TIP.2015.2481326.
[14] X. Li, S. Luo, and J. Li, “Liver segmentation from CT image using fuzzy clustering and level set,” Journal of Signal and
Information Processing, vol. 04, no. 03, pp. 36–42, 2013, doi: 10.4236/jsip.2013.43B007.
[15] N. Sharma et al., “Automated medical image segmentation techniques,” Journal of Medical Physics, vol. 35, no. 1, p. 3, 2010,
doi: 10.4103/0971-6203.58777.
[16] Y. Chen, Z. Wang, J. Hu, W. Zhao, and Q. Wu, “The domain knowledge based graph-cut model for liver CT segmentation,”
Biomedical Signal Processing and Control, vol. 7, no. 6, pp. 591–598, Nov. 2012, doi: 10.1016/j.bspc.2012.04.005.
[17] T. Fan, G. Wang, X. Wang, Y. Li, and H. Wang, “MSN-Net: a multi-scale context nested U-Net for liver segmentation,” Signal,
Image and Video Processing, vol. 15, no. 6, pp. 1089–1097, Sep. 2021, doi: 10.1007/s11760-020-01835-9.
[18] Z. Bai, H. Jiang, S. Li, and Y.-D. Yao, “Liver tumor segmentation based on multi-scale candidate generation and fractal residual
network,” IEEE Access, vol. 7, pp. 82122–82133, 2019, doi: 10.1109/ACCESS.2019.2923218.
[19] X. Li, H. Chen, X. Qi, Q. Dou, C.-W. Fu, and P.-A. Heng, “H-DenseUNet: hybrid densely connected UNet for liver and tumor
segmentation from CT volumes,” IEEE Transactions on Medical Imaging, vol. 37, no. 12, pp. 2663–2674, Dec. 2018, doi:
10.1109/TMI.2018.2845918.
[20] Z. Wang, Y. Zou, and P. X. Liu, “Hybrid dilation and attention residual U-Net for medical image segmentation,” Computers in
Biology and Medicine, vol. 134, p. 104449, Jul. 2021, doi: 10.1016/j.compbiomed.2021.104449.
[21] M. Lawankar, S. Sangewar, and S. Gugulothu, “Segmentation of liver using marker watershed transform algorithm for CT scan
images,” in 2016 International Conference on Communication and Signal Processing (ICCSP), Apr. 2016, pp. 0553–0556, doi:
10.1109/ICCSP.2016.7754200.
[22] Y. M. Y. Abdallah, “An accurate liver segmentation method using parallel computing algorithm,” Journal of Biomedical
Engineering and Medical Imaging, vol. 2, no. 3, pp. 15–23, Jun. 2015, doi: 10.14738/jbemi.23.1187.
[23] H. T. Huynh, N. Le-Trong, P. T. Bao, A. Oto, and K. Suzuki, “Fully automated MR liver volumetry using watershed
segmentation coupled with active contouring,” International Journal of Computer Assisted Radiology and Surgery, vol. 12, no. 2,
pp. 235–243, Feb. 2017, doi: 10.1007/s11548-016-1498-9.
[24] A. M. Anter and A. E. Hassenian, “CT liver tumor segmentation hybrid approach using neutrosophic sets, fast fuzzy c-means and
adaptive watershed algorithm,” Artificial Intelligence in Medicine, vol. 97, pp. 105–117, Jun. 2019, doi:
10.1016/j.artmed.2018.11.007.
[25] M. Ahmad et al., “A lightweight convolutional neural network model for liver segmentation in medical diagnosis,”
Computational Intelligence and Neuroscience, vol. 2022, pp. 1–16, Mar. 2022, doi: 10.1155/2022/7954333.
[26] W. Tang, D. Zou, S. Yang, J. Shi, J. Dan, and G. Song, “A two-stage approach for automatic liver segmentation with faster R-
CNN and DeepLab,” Neural Computing and Applications, vol. 32, no. 11, pp. 6769–6778, Jun. 2020, doi: 10.1007/s00521-019-
04700-0.
[27] S. Di, Y. Zhao, M. Liao, Z. Yang, and Y. Zeng, “Automatic liver tumor segmentation from CT images using hierarchical iterative
superpixels and local statistical features,” Expert Systems with Applications, vol. 203, Oct. 2022, doi:
10.1016/j.eswa.2022.117347.
[28] P. Bilic et al., “The liver tumor segmentation benchmark (LiTS),” arXiv preprint arXiv:1901.04056, Jan. 2019.
[29] S. S. Nidhyanandhan, R. Sindhuja, and R. S. S. Kumari, “Double stage gaussian filter for better underwater image enhancement,”
Wireless Personal Communications, vol. 114, no. 4, pp. 2909–2921, Oct. 2020, doi: 10.1007/s11277-020-07509-6.
[30] R. A. Haddad and A. N. Akansu, “A class of fast Gaussian binomial filters for speech and image processing,” IEEE Transactions
on Signal Processing, vol. 39, no. 3, pp. 723–727, Mar. 1991, doi: 10.1109/78.80892.
[31] H. M. El-Hoseny et al., “An optimal wavelet-based multi-modality medical image fusion approach based on modified central
force optimization and histogram matching,” Multimedia Tools and Applications, vol. 78, no. 18, pp. 26373–26397, Sep. 2019,
doi: 10.1007/s11042-019-7552-1.
[32] M. Nixon and A. Aguado, Feature extraction and image processing for computer vision. Academic press, 2019.

1549
[33] J. Yousefi, “Image binarization using Otsu thresholding algorithm,” Ontario, Canada: University of Guelph, pp. 1–4, 2011, doi:
10.13140/RG.2.1.4758.9284.
[34] N. Zhu, G. Wang, G. Yang, and W. Dai, “A fast 2D Otsu thresholding algorithm based on improved histogram,” in 2009 Chinese
Conference on Pattern Recognition, Nov. 2009, pp. 1–5, doi: 10.1109/CCPR.2009.5344078.
[35] J. G. M. Schavemaker, M. J. T. Reinders, J. J. Gerbrands, and E. Backer, “Image sharpening by morphological filtering,” Pattern
Recognition, vol. 33, no. 6, pp. 997–1012, Jun. 2000, doi: 10.1016/S0031-3203(99)00160-0.
[36] R. M. Haralick and L. G. Shapiro, Computer and robot vision, vol. 1. Addison-wesley Reading, 1992.
[37] P. P. Acharjya, A. Sinha, S. Sarkar, S. Dey, and S. Ghosh, “A new approach of watershed algorithm using distance transform
applied to image segmentation,” International Journal of Innovative Research in Computer and Communication Engineering,
vol. 1, no. 2, pp. 185–189, 2013.
[38] J. B. T. M. Roerdink and A. Meijster, “The watershed transform: Definitions, algorithms and parallelization strategies,”
Fundamenta Informaticae, vol. 41, no. 1,2, pp. 187–228, 2000, doi: 10.3233/FI-2000-411207.
[39] R. Beare and G. Lehmann, “The watershed transform in ITK - discussion and new developments,” The Insight Journal, Jun. 2006,
doi: 10.54294/lf8u75.
[40] A. Kornilov and I. Safonov, “An overview of watershed algorithm implementations in open source libraries,” Journal of Imaging,
vol. 4, no. 10, Oct. 2018, doi: 10.3390/jimaging4100123.
[41] V. J. Gaikwad, “Marker-controlled watershed transform in digital mammogram segmentation,” International Journal for
Research in Applied Science & Engineering, vol. 3, no. 3, pp. 18–21, 2015.
BIOGRAPHIES OF AUTHORS
Kiran Malhari Napte received the B.Eng. degree in electronics engineering from
Pune University, India in 2010 and the Master of Engineering degrees in Electronics &
Telecommunication from Pune University in 2014. Currently, he is an Assistant Professor at
the Department of Electronics and Telecommunication Engineering, PimpriChinchwad
College of Engineering and Research, Ravet, Pune. His research interests include medical
image processing, signal processing, pattern recognition, power electronics and artificial
intelligence in healthcare. He can be contacted at email: kiran.napte@pccoer.in.
Anurag Mahajan received B.E. in Electronics Engineering from Vikram
University, Ujjain in 1998. He received M.E. in Electronics and Communication Engineering
from University of Roorkee, Roorkee (Now IIT, Roorkee) in 2000. He received Ph.D. from
Jaypee University of Engineering and Technology, Guna in 2014.He worked in Medi-Caps
Institute of Technology and Management, Indore in Electronics and Communication
Engineering Department from August 2002 to January 2008 at different position from Lecturer
to Reader. He worked as Assistant Professor in Jaypee University of Engineering and
Technology, Guna in Electronics and Communication Engineering Department from January
2008 to December 2017. Presently he is working as Assistant Professor in Electronics and
Telecommunication Engineering Department in Symbiosis Institute of Technology, Symbiosis
International (Deemed University), Pune since December 2017. He can be contacted at email:
anurag.mahajan@sitpune.edu.in.

Liver segmentation using marker controlled watershed transform

More Related Content

Similar to Liver segmentation using marker controlled watershed transform (20)

More from IJECEIAES (20)

Recently uploaded (20)

Liver segmentation using marker controlled watershed transform