Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications

Volume: 05 Issue: 02 December 2016, Page No.172-177

ISSN: 2278-2419

Image Segmentation Based Survey on the Lung Cancer

MRI Images
S. Perumal1, T. Velmurugan2
1
Research scholar,Department of Computer Science,Vels University, Chennai, India.
2
Associate Professor,PG and Research Dept. of Computer Science, D.G.Vaishanav College, Chennai, India.
Mail: [email protected]; [email protected]

Abstract: Differentiating cancer affected part in lungs and image and processing of image. Various methodologies are
giving proper solution to the problem are toughest job in carried out to extract the cancer affected part in the lung. This
medical field. Doctors face many problems in correctly research work discusses about the different research articles
spotting up of cancer affected area in lungs. Image processing that are carried out by various researchers in the literature.
can be a solvent for this type of issues, especially to identify The organization of the paper is structured as follows. Session
the cancer affected areas in lungs. Historical data of different 2 discusses about introduction and some of the related papers
types of lung cancer images are collected and image processing of image segmentation. Session 3 explores the lung cancer
methods are carried out for the identification of cancer affected image segmentation in detail. Finally, section 4 concludes the
regions in the lungs by the physicians and experts. This research work based on the results of related works of the
research work carried out a survey on the lung cancer data previous papers.
analysis done by various researchers. Also, it suggests the best
method and technique applied for the prediction of cancer in II. IMAGE SEGMENTATION TECHNIQUES
the affected parts of lungs.
Image processing follows various steps in processing cancer
Keywords: Image Segmentation, Lung Cancer Data, Image affected parts of lungs. Image segmentation plays a major role
Clustering, k-Means Clustering, Fuzzy C-Means Clustering for every image processing process. In image segmentation
process, the digitally converted lung image has to be divided
I. INTRODUCTION into various regions. Usually an image segmentation technique
identifies objects, sets boundary and identifies other relevant
Image processing is one of the core area used for various regions in the image. This survey focuses on various steps
domains to identify cancer affected regions in lungs of MRI involved in images segmentation and researchers work on lung
images. Identification of cancer affected parts in lung is mainly cancer images. The image segmentation can be categorized
initiated with image processing techniques such as noise into six major types, which is very clearly shown below.
removal, feature extraction, identification of affected regions
and possible comparison with historical data of lung cancer.
Usually, digital image processing follows many techniques to
unite different shapes in image into single unit. In this article, it
is followed with clever bit technique for identifying particular
region in the lung image. The region identified through the
segmentation technique can be viewed from different angles
and with different lighting. The basic advantages in choosing
the technique is to identify, color difference between cancers
affected region and those not affected parts by finding the
intensity of images.

The human visual system performs special tasks for identifying

color difference between cancer affected area and not affected Figure 1: Image Segmentation types
area in the lung image. Very skilful programming and lots of
processing power are necessary in identification of color II.1.Edge Image Segmentation
difference. Manipulations of data in the form of an image Edge or Corner boundary is a part or region between two
through several possible techniques are very essential for different regions, mostly defined with gray scale level.
identification of color through image segmentation techniques. Detection or identification of corners is carried out with
An image is usually interpreted as a two-dimensional array of identified object borders which are surrounded, which also
brightness values, and is most familiarly represented by such helps in detecting intensity value of an image. Corner
patterns as those of a photographic print, slide, television identification technique is the most essential part for analyzing
screen, or movie screen. An image can be processed optically an image and recognition of patterns. Corner identification
or digitally with a computer. Image processing has the provides a physical extent to the object and follows different
combination of various artificial intelligence areas like fuzzy detection methodologies as follows:
logic, pattern recognition and machine learning. The essential
technique, (Image processing) is carried out for segmenting the Roberts Edge Detection: Roberts’s Edge detection is one of
images and used for further process. The image segmentation the oldest image processing edge detection technique stated by
layers can be differentiated as understanding image, analysis Lawerence Roberts in the year 1963. The edge detection
1
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 05 Issue: 02 December 2016, Page No.172-177
ISSN: 2278-2419
operator used by Roberts is very simple, quick to execute and background. In this method, a grey scale image is converted
very easy to measure 2-D spatial gradient of an image. The into binary image. The binary image contains all the necessary
operator also gives highlighted regions for high spatial gradient data regarding location and shape of the objects. Conversion to
regions which often correspond to edges. The Roberts operator binary image is useful because it reduces the complexity of
uses grayscale image for processing and provides the resultant data. Threshold methods are as follows:
images in grayscale. Absolute magnitude spatial gradient
points are calculated from the output image pixel [1]. Global Thresholding: In the global thresholding, the intensity
value of the input image should have two peak values which
Sobel Edge Detection: Sobel edge detection is referred with correspond to the signals from background and objects. It
the author Irwin Sobel and it is also called as Sobel filter. It has shows the degree of intensity separation between two peaks in
two masks, in which one is horizontal and other one is vertical. an image.
Generally 3*3 metric masks are used. Standard Sobel
operators, for a 3×3 neighborhood, each simple central Variable Thresholding: In variable thresholding, we separate
gradient estimate is vector sum of a pair of orthogonal vectors. the foreground image objects from the background, based on
Each orthogonal vector is a directional derivative estimate the intensities of each region.
multiplied by a unit vector specifying the derivative’s  Local or regional thresholding.
direction. The vector sum of these simple gradient estimates  Adaptive thresholding.
amounts to a vector sum of the 8 directional derivative vectors.
Thus for a point on Cartesian grid and its eight neighbors Multiple Thresholding: Multiple thresholding can be defined
having density values as shown: [2] as that segments a grey level image into several distinct regions
[9]. . It defines more than one threshold for the given image
Prewitt Edge Detection: Corner detection or Edge detection and divides the image into certain brightness regions and it
algorithm is most popular Prewitt Detector technique used in corresponds to the background and several objects.
image processing. It is also known as Discrete Differentiation
operator. Gradient of the image intensity function is calculated Artificial Neural Network Image Segmentation
by using it. Applying a horizontal and vertical filter in Wencang Zhao [10] proposed a new image segmentation
sequence Prewitt Edge filter detects edges. These two basic algorithm based on textural features[11] and Neural
convolution filters are applied to the image and summed to Network[12] to separate the targeted images from background.
form the final result [3]. Dataset of micro-CT images are used. De-noising filter is used
to remove noise from image as a pre-processing step, Feature
Fuzzy Image Segmentation : Liu Yucheng [4] introduced extraction is performed next, and then Back Propagation
fuzzy image segmentation algorithm which uses morphological Neural Network is created, and finally, it modifies the weight
opening and closing operations to enhance the image and number of network, and save the output. Proposed algorithm is
perform the gradient operations on the resultant image [5]. compared with Thresholding method and Region Growing
Then compare the fusion algorithm with Watershed algorithm method. Based on speed and accuracy of segmentation, results
[6] and Prewitt methods, which found that fusion approach shows proposed technique outperforms other methods .Lijun
gives solution to the problem of over-segmentation of Zhang [13] proposed image segmentation based on neural
Watershed algorithm. It also saves the information about network system with color images. They combined the
image and also improves the speed. Wavelet Decomposition and Self Organizing Map (SOM) to
propose a new method, i.e., SOM-NN. Voting among child
Syoji Kobashi [7] used fuzzy image segmentation (scale based) pixels selected the parent pixel. After initialization, ANN
and fuzzy object model to segment the cerebral parenchyma found the segmentation result which satisfies all levels.
region of new born brain MRI image. In first step Foreground Wavelet decomposition is performed to remove noise. Hence
region is separated, next correction of MRI intensity in- wavelet decomposition and SOM-NN are combined to perform
homogeneity is performed and then scale-base Fuzzy Object segmentation. Results have shown that the used methods has
Model (FOM) is applied on resultant image. Results of reduce noise and produce accurate segmentation. Shohel Ali
proposed method are evaluated on the basis of Fast Positive Ahmed [14] proposed Image Texture Classification technique
Volume Fraction (FPVF) and Fast Negative Volume Fraction based on Artificial Neural Networks (ANN). Initially, image is
(FVNF). Results from experiment have shown that FOM captured and pre-processing is performed, then feature
(Fuzzy object model) has attained minimum FPVF and FVNF extraction process is carried out [15] is performed, whereas,
values. RefikSamet [8] proposed a new Fuzzy Rule based ANN classifier [16] is used for texture classification,
image segmentation technique to segment the rock thin Clustering is performed to separates background from sub-
segment images. RGB image of rock thin segment is taken as images. Trained ANN combines the input pixels into two
input and segmented mineral image is given as output. Fuzzy C clusters which gives results with texture classification and
Means is also applied on rock thin images and results are segmentation of image as a ANN product.
compared with both techniques. The researchers take sample
image from minerals and features are distinguished on the II.3.Region Based Image Segmentation
basis of red, green and blue components of image. Region Based segmentation is defined as a segment that
produces similar image into various regions. It is used to
II.2.Threshold Image Segmentation determine the region directly by using grey values of the image
In image segmentation, Threshold method technique is widely pixels and portioning is done. Two basic techniques of region
used. It is mainly used to discriminate foreground from based segmentation are given below.
2
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 05 Issue: 02 December 2016, Page No.172-177
ISSN: 2278-2419
Region Growing Methods: Region growing is a technique accurately. Chun Yuan[20] proposed a new segmentation
that groups pixels or sub regions into larger regions based on model for color images. This model is based on Geodesic
predefined criteria. The pixels aggregation starts with a set of Active Contour (GAC) model. But GAC is only restricted to
seed points in a way that the corresponding regions grow by gray scale images. Therefore this model has an extension of
appending to each seed points those neighboring pixels that GAC model, and known as color-GAC model. It uses the
have similar properties like grey scale, color, texture, shape expression of the Gradient of color image.
etc. [17]
II.5.Clustering Based Segmentation
Region Splitting and Merging: In case of region splitting, the Clustering based image segmentation segments images at grey
whole image is taken as a single region and then this region is level. It is easy to apply directly and can be extended to high
being break into a set of disjoint regions which are coherent dimension. Clustering is also applicable in multispectral and
with themselves. Region merging opposes Region Splitting. A images in color.
merging technique is used after each split and compares
adjacent regions and merges them. It starts with small regions III. IMAGE SEGMENTATION OF LUNG CANCER
and merge the regions which have similar characteristics like DATA
grayscale, variance etc. Nikhil R Pal [21], Edge Detection Method displays gray scale
images depends on discontinuity detection, generally with less
II.4.Partial Differential Equation (PDE) Image or more rapid gray level changes. This approach works with
Jinsheng Xiao [18] proposed a new non-linear discontinue human predictive objects and aims to findthe excellent region
partial differential equation (PDE) that models the level set disparity in image processing. V.K. Dehariya [22] uses the
method of gray images. A discrete method is also proposed to various peaks in images with corresponding regions.
find numerical solution and to implement the filter.Non-Linear Computational time and image complexity for given image are
discontinue PDE formula is applied on image of cameramen reduced by applying this method. Rastgarpour M [23]
using MATLAB. Results have shown that image edges and assembles pixels in uniform regions and counts the region
boundaries are remained blurred and can be shifted by using growth, split value, and merging their permutation. This
Close operator. More information can be saved by using the method works well if homogeneity region norm in plain less
proposed scheme. Fengchun Zhang [19] presents a variation image and removes irrelevance in image. Image noise removal
model using 4th order PDE with 2nd order PDE for finger vein technique followed in this paper is comparably stronger than
image de-noising. Midpoint Threshold segmentation technique Edge Detection method. Yang Yang [24] uses fuzzy operators,
is used to extract the region of interest accurately. Fourth order mathematics, properties and inference rules and give a mode to
PDE has reduced the noise very well, whereas 2nd order PDE handle the uncertainty inheritance range in troubles, because of
has approximated the boundaries effectively. It can be ambiguity instead of randomness. T.F. Wang [25] extends
observed from experiments that PSNR value of proposed training period in working with image and finds that the
method increases by 2 dB. This method is compared with initialization might affect the outcome problem as shown in
threshold based segmentation algorithm and it is found that Table 1.
proposed method has segment the real finger vein image

Table 1: Image segmentation of lung cancer data

Authors Segmentation Description of Method Benefits of Method Findings/ Results

Method

Nikhil R Pal et.al. Edge Displays gray scale image Human preferred Edges are not clearly define
[21] Detection level changes objects can be Very difficult to form a
Method implemented with boundary for images, but
fine region irrelevance in image can be
arrangements. removed.

Xu, Lei, et. al. [26] Edge Initially a straight line is Randomly pick two A curve expressed by a n
Detection taken and Randomized or n pixels and map parameter equation, instead of
Method Hough Transform (RHT) them into one points transforming one pixel into an
procedure is implemented on in the parameter. n-1 dimensional hyper surface
various curves. in the parameter space.
V. K. Dehariya et.al. Thresholding Image peaks are corresponds Image knowledge is Doesn’t work properly in nice
[22] Method to particular regions in not needed and image peak and plane images.
image. computational time No continues segmentation of
is reduced. regions is performed.
W. Jentzen., et al. Thresholding Optimum threshold obtained These findings are Phantom data analysis showed
[27] Method from the adaptive associated with the that the S/B-threshold-volume
thresholding method requires limitation of the curves of 18F-FDG and 124I
a priori estimation of the ITM. The ITM is were similar.
lesion volume from anatomic especially useful for
3
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 05 Issue: 02 December 2016, Page No.172-177
ISSN: 2278-2419
images such as CT. lesions that are only
visible on PET.
Rastgarpour M. et.al. Region Separating, Counting, More effective noise Deals with computational time
[23] Dependent Merging and assembling detection technique and segmentation based on
Method pixel are performed. than Edge Detection splitting emerge pixel.
method
U. Nestle, et. al. [28] Region Data on the primary tumors The different We found substantial
Dependent of 25 patients with NSCLC techniques of tumor differences between the 4
Method were analyzed. They had F- contour definition methods of up to 41% of the
FDG PET during initial by F-FDG PET in GTV vis. The differences
tumor staging. radiotherapy correlated with SUV max,
planning lead to tumor homogeneity, and
substantially lesion size.
different volumes,
especially in
patients with
inhomogeneous
tumors
Yang Yanget.al.[24] Fuzzy Fuzzy operators, Fuzzy membership Calculation occupied in fuzzy
Method mathematics and function could be approaches could be intensive
implementation rules utilized to show the
degree of few
properties
Brown, M.S et. al. Fuzzy Nodules are detected from An automated Images from 15 subjects that
[29] Method among the opacities, modeled system was were not part of the
as inside the lung. developed for development set were used to
Morphologic closing is detecting lung micro test the system. Among these
applied to the lung so that nodules on thin- images, the truth committee
nodules that contact the chest section computed determined that there were
wall are enclosed. topographic images collectively 57 micro nodules
and was applied to smaller than 3 mm and 22
data from 15 nodules larger than 3 mm
subjects with 77
lung nodules.
Hill, D. L., et. al [30] Fuzzy The word registration is used Many registration The classical Procrustes
Method with two slightly different algorithms involve problem, i.e. T ∈ {rigid body
meanings. The first meaning iteratively transformations} has known
is determining a transforming image solutions. A matrix
transformation that can relate A with respect to representation of the rotational
the position of features in one image B while part can be computed using
image or coordinate space optimizing a singular-value decomposition
with the position of the similarity measure (SVD)
corresponding feature in calculated from the
another image or coordinate voxel values.
space.
Kuhnigk, J. M., et. al Fuzzy The exact localization of the Lung segmentation The preliminary intra- and
[31] Method lobe-separating fissures in procedure is fully inter-observer studies
CT images often represents a automated and conducted already indicate a
non-trivial task even for therefore does not low variability (Similarity >
experts. Therefore, a lung add to the user input 99.5%) of the generated
lobe segmentation method necessary for the results.
suitable to work robustly lobar segmentation.
under clinical conditions Since other methods
must take advantage of exist to fulfill this
additional anatomic task and we would
information. rather focus on the
lobe segmentation.
T.F. Wang et.al.[25] Neural Classification, Clustering and Could entirely Initialization might affect the
Network Neural Network are used. exploit the parallel outcome.
Method nature of neural net.

4
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 05 Issue: 02 December 2016, Page No.172-177
ISSN: 2278-2419
IV. CONCLUSION algorithm," in Proc. Sixth International Conference on
Image segmentation is the most important part of image Semantics Knowledge and Grid (SKG), pp. 395-397,
processing techniques. This paper shows a clear cut idea for 2010.
researchers to carry out research in image segmentation. This [13] Ahmed, S. A., Dey, A., and Sarma, K. K., "Image texture
article also focuses on existing work carried out in this broad classification using Artificial Neural Network (ANN)," in
area of image segmentation. It also provides lot of ideas about Proc. 2nd National Conference on Emerging Trends and
measurements and methods used by the image segmentation Applications in Computer Science (NCETACS), pp. 1-4,
researchers with description of each and every researchers 2011.
view. Hence, this work concludes that most of the researchers [14] Sharif, M., Raza, M., Mohsin, S., and Shah, J. H.,
states fuzzy approaches give better results for any kind of "Microscopic feature extraction method," Int. J. Advanced
image processing data. Networking and Applications, vol. 4, pp. 1700-1703,
2013.
Reference [15] Irum, I., Raza, M., and Sharif, M., "Morphological
[1] Zhang,Y.J, “An Overview of Image and Video techniques for medical images: A review," Research
Segmentation in the last 40 years”, Proceedings of the 6th Journal of Applied Sciences, vol. 4, pp.1-5, 2012.
International Symposium on Signal Processing and its [16] Xiao, J., Yi, B., Xu, L., and Xie, H., "An image
Applications, pp. 144-151, 2001. segmentation algorithm based on level set using
[2] SOBEL, I., “An Isotropic 3×3 Gradient Operator, Machine discontinue PDE," in Proc. First International Conference
Vision for Three – Dimensional Scenes”, Freeman, H., on Intelligent Networks and Intelligent Systems,
Academic Press, NY, pp. 376-379, 1990. ICINIS'08., pp. 503-506, 2008.
[3] Yucheng, L., and Yubin, L., "An algorithm of image [17] Zhang, F., Guo, S., and Qian, X., "Segmentation for finger
segmentation based on fuzzy mathematical morphology," vein image based on PDEs denoising," in Proc. 3rd
International Forum on Information Technology and International Conference on Biomedical Engineering and
Applications, IFITA'09, pp. 517-520, 2009. Informatics (BMEI), pp. 531-535, 2010.
[4] Haider, W., Sharif, M and Raza, M., "Achieving accuracy [18] Yuan, C., and Liang, S., "Segmentation of color image
in early stage tumor identification systems based on image based on partial differential equations," in Proc. Fourth
segmentation and 3D structure analysis," Computer International Symposium on Computational Intelligence
Engineering and Intelligent Systems, vol. 2, pp. 96-102, and Design (ISCID), pp. 238-240, 2011.
2011. [19] Kaur, A. (2012). A Review Paper on Image Segmentation
[5] Shahzad, M. S. A., Raza, M., and Hussain, K., "Enhanced and its Various Techniques in Image
watershed image processing segmentation," Journal of Processing. International Journal of Science and Research.
Information and Communication Technology, vol. 2, pp. 2012.
1-9, 2008. [20] Huang, L. Y., Chen. D. R., “Watershed segmentation for
[6] Kobashi, S., and Udupa, J. k., "Fuzzy object model based breast tumor in 2D sonography”, Ultrasound in Medicine
fuzzy connectedness image segmentation of newborn & Biology, vol:30(5), pp. 625-632, 2004.
brain MRI images," in Proc. IEEE International [21] Nikhil R Pal and Sankar K Pal, “A review on Image
Conference on Systems, Man, and Cybernetics (SMC), pp. Segmentation Techniques”, Indian Statistical Institute,
1422-1427, 2012. Pattern Recognition, Vol 26 (9), pp.1277, 1993.
[7] Samet, R., Amrahov, S. E., and Ziroglu, A. H., "Fuzzy [22] Dehariya, V. K., Shrivastava, S. K., Jain, R. C.,
rule-based image segmentation technique for rock thin “Clustering of Image Data Set Using K-Means and Fuzzy
section images," in Proc. 3rd International Conference on K-Means Algorithms”, International conference on CICN,
Image Processing Theory, Tools and Applications (IPTA), pp. 386- 391, 2010.
pp. 402-406, 2012. [23] Rastgarpour M., and Shanbehzadeh J., “Application of AI
[8] Zhao, W., Zhang, J., Li, P., and Li, Y "Study of image Techniques in Medical Image Segmentation and Novel
segmentation algorithm based on textural features and Categorization of Available Methods and Tools”,
neural network," in International Conference on Intelligent Proceedings of the International Multi Conference of
Computing and Cognitive Informatics (ICICCI), pp. 300- Engineers and Computer Scientists, IMECS, Vol 1(1),
303, 2010. 2011.
[9] Velmurugan, T., and Sukassini, T. P., “Segmentation of [24] Yang Yang, Yi Yang, Heng Tao Shen, Yanchun Zhang,
Mammogram Image using Multilevel Threshold and Xiaoyong Du, Xiaofang Zhou, “ Discriminative
Gravitational Search Algorithm”, International Journal of nonnegative spectral clustering with out of sample
Computer Technology and Applications, Vol: 9(40), pp. extension ” IEEE Transactions on knowledge and data
67-75, 2016. engineering, Vol. 25(8), pp:1760-1771, 2013.
[10] Sharif, M., Javed, M. Y., and Mohsin, S., "Face [25] Wang, T. F., and Li, D. Y “Automatic segmentation of
recognition based on facial features," Research Journal of medical ultrasound image using self-creating and
Applied Sciences, Engineering and Technology, vol. 4, pp. organizing neural network”, Journal of electronics, Vol. 21
2879-2886, 2012. (1), pp.124-127. 1999.
[11] Yasmin, M., Sharif, M., and Mohsin, S., "Neural networks [26] Xu, Lei, Erkki Oja, and Pekka Kultanen. "A new curve
in medical imaging applications: A survey," World detection method: randomized Hough transform
Applied Sciences Journal, vol. 22, pp. 85-96, 2013. (RHT)." Pattern recognition letters, Vol. 11(5), pp. 331-
[12] Zhang, L., and Deng, X., "The research of image 338, 1990.
segmentation based on improved neural network
5
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 05 Issue: 02 December 2016, Page No.172-177
ISSN: 2278-2419
[27] Jentzen, W., Freudenberg, L., Eising, E. G., Heinze, M.,
Brandau, W., and Bockisch, “A., Segmentation of PET
volumes by iterative image thresholding”, Journal of
Nuclear Medicine, Vol. 48(1), pp. 108-114.2007.
[28] Nestle, U., Kremp, S., Schaefer-Schuler, A., Sebastian-
Welsch, C., Hellwig, D., Rübe, C and Kirsch, C. M.,
"Comparison of different methods for delineation of 18F-
FDG PET–positive tissue for target volume definition in
radiotherapy of patients with non–small cell lung
cancer." Journal of Nuclear Medicine, Vol. 46(8), pp.
1342-1348. 2005.
[29] Brown, M. S., Goldin, J. G., Suh, R. D., McNitt-Gray,
M.F., Sayre, J.W. and Aberle, D.R., “Lung micro nodules:
automated method for detection at thin-section CT—initial
experience”. Vol .226(1), pp.256-262. 2003.
[30] Hill, D.L.,Batchelor,P.G., Holden,M., & Hawkes,D.J,
“Medical image registration”. Physics in medicine and
biology, Vol: 46(3), pp. 1-5, 2001.
[31] Kuhnigk, J. M., Hahn, H., Hindennach, M., Dicken, V.,
Krass, S., & Peitgen, H. O. “Lung lobe segmentation by
anatomy-guided 3 D watershed transform”. In Proceedings
of SPIE, Vol. 5032, pp. 1482-1490, 2003.