Professional Documents
Culture Documents
ABSTRACT:
From last few years lung cancer become sign of fear among the people all over the world. As a
result many countries generate funds and give invitation to many scholars to overcome on this
disease. Many researchers proposed many solutions and challenges of different phases of
computer aided system to detect the lung cancer in early stages and give the facts about the lung
cancer. Computer vision play vital role to prevent lung cancer. Since image processing is
necessary for computer vision, further in medical image processing there are many technical
steps which are necessary to improve the performance of medical diagnostic machines. Without
such steps programmer is unable to achieve accuracy given by another author using specific
algorithm or technique. So in this review paper we highlight such steps which are used by many
author in pre-processing, segmentation and classification methods of lung cancer area detection.
If pre-processing and segmentation process have some ambiguity than ultimately it effects on
classification process. We discuss such factors briefly so that new researchers can easily
understand the situation to work further in which direction.
process. The separation of a channel from there are two strategies found in various
other channels is called layer separation, literature such as edge based segmentation
which is key step to produce the satisfactory and region based segmentation. Region
results [11]. based segmentation is further divided as
Thresholding,
2.7 GRAY CONVERSION. Region growing
Medical image datasets are found in Clustering in feature space
number of different formats. For the
simplicity we divide images in three 3.2 THRESHOLDING.
following types A value which is able to segment the ROI
i) Color image is called threshold value. In lung cancer case
ii) Gray image we convert our image into gray image and
iii) Binary image then decide the threshold value on the basis
Color image has three channels each of tumor properties. For example we found
channel has 256 variant intensity values that tumor has intensity values 115 to 255
ranging 0 to 255. In RGB if red, green and then we can convert all values which are less
blue have zero value then color is black than 115 into 0 and all other values between
while all channel values have 255 then color 115 to 255 is equal to 1 in binary image. In
is white. Computer can easily distinguish this way all unwanted area become black
each color shade while human vision is and wanted area become white region [19].
limited as compare to machine vision. In
color image there are almost 16777216 3.3 Region Growing
different shade. In gray image there are 256 In region growing method we group the
shades in the form of black and white where pixels or subparts of the image into the large
0 represents black color and 256 represents region, means to say we analysis the pixel
white color value and then we combine the various
In the other hand, binary image has only two intensity values which are similar properties
shades. 0 is used for Black and 1 is used for such as color, gray level values, texture and
white. Usually we convert color image to shape features. While in region splitting
gray image for extracting image objects method we take large image and then split it
from the whole image by adjusting into small region on the basis of
histogram of the image [17]. homogeneity of that region. [20].
image [21]. Clustering techniques are possible sub-image with in the whole image.
considered as unsupervised classification This process is continue until the template is
methods in which k-means clustering exactly matched in term of intensity values
algorithm is very famous in machine with any portion of an image [13]. There are
learning techniques where the output is not two types of matching Local Matching and
given in the features vector [2]. 3D Elastic Matching. The Local Matching
method based on GGVF (Generalized
3.5 EDGE BASED SEGMENTATION Gradient Vector Flaw). One of the GGVF’s
METHODS. features is to extend the influence of
In edge based segmentation we try to find gradient also to position away from the edge
edges of digital objects found in an image on and a homogeneous region while leaving
the basis of abruptly change found in desirable properties of vector field on the
intensity values of a pixel. Many digital edge map. The 3D Elastic matching method
objects has cloudy and broken edges To based on smoothing shift vector. This
remove such effects from the image we method is minimization of energy function
usually performs morphological operations by sequentially updating the shift vector on
which are discussed in next section and the alignment points from initial vector. This
linear filters to give more distance and initial vector is taken from shift vector
remove distance which is closed to the obtained from local matching.
digital object [12]. Because tumor has no 3.8 MARKER-CONTROLLER
regular shape and intensity range so it is SEGMENTATION.
challenging task for Computer Aided Design To reduce the drawback of over
(CAD). segmentation in watershed algorithm, many
researcher use marker controller
3.6 MORPHOLOGICAL segmentation in which there are two types of
SEGMENTATION. markers, one is internal marker and other is
Morphological base segmentation is a external marker. In internal marker we give
process that combine morphological the range of gray level values of required
operations such as extended minima and region while in external marker we give the
morphological gradient with watershed range of gray values related to back ground
flooding algorithm to segment grayscale of the image [24]. The marker selection
image of any type of image. For example, typically consists of two steps:
gray scale with single band (8 bit image) and preprocessing and definition of a set of
color image with three bands (24 bit image). criteria that markers must satisfy. The
In morphological operation, popular preprocessing scheme is to filter an image
operations are Dilation and Erosion. Erosion with a smoothing filter. This step can
operation is performed when two edges are minimize the effect of small spatial detail, in
merged with each another whereas Dilation other words, this step is to reduce the large
operation is performed when there is broken number of irrelevant detail, which is the
edges are found in the image This operation reason of over-segmentation. Many
try to fill the gap between two broken edges researcher used marker controller
[22]. segmentation in lung cancer detection in
medical imaging.
3.7 TEMPLATE MATCHING.
In template matching method we take a 4.1 FEATURE EXTRACTION.
sample or template and match it with all
Mehran University Research Journal of Engineering and Technology
Dated: 28 Apr 2018
Each well define object has some features approach is based on the number of black
or properties which helps us to identify that pixels which are greater than white pixels if
object easily. To automatically classify the not then it means image is not normal.
objects we need some features. Collection of
these features are called features vector. For 4.3 MASKING
the identification of lung cancer we need to Masking is also called filtering the word
be find such features from the image. In “filtering” has been borrowed from the
image processing there are three types of frequency domain. Filters are classified as:
features, first is structural features, second is
statistical texture features and third is Low-pass (i.e., preserve low frequencies)
spectral features. Structural features are also High-pass (i.e., preserve high frequencies)
called binary features like area, centroid, Band-pass (i.e., preserve frequencies within
perimeter, orientation, projection, aspect a band)
ratio and Euler number etc. Statistical Band-reject (i.e., reject frequencies within a
features also divide two classes’ first order band) [1].
and second order statistical texture features. In masking approach we take usually
First order features are extracted directly neighborhood pixels of CT scan images of
from the gray level histogram while in lung cancer with effected areas and then
second order texture features first we find calculate the statistical and structural
co-occurrence matrix and then we find features which are used to classify Lung
mean, entropy, and co-variance. In spectral cancer images in early stage and normal
features Gaber and wavelet features are very images as well.
popular features. Many authors take
different set of features to detect the lung 5.1 CLASSIFICATION
cancer with different methodology with Human brain is a great example of ability to
different available dataset but still there is learn and classify the different objects and
no any method which can automatically things without any error. For example,
segment and classify the malignant and distinctive mankind's face and pictures. At
benign area in the Lung cancer image. the same time it still challengeable task for
However there are two approaches are used the human to get the ability via machines.
to extract the features vector first However there are lot of techniques have
Banalization approach related to structural found in literature for data classification. But
features which are calculated from binary no single classification technique has been
image and other is masking approach in found with best accuracy for all kinds of
which we usually extract statistical texture dataset [25]. Two main approaches are used
and spectral features. to make a machine capable to understand
and classify the CT scan images are
4.2 BANALIZATION supervised and unsupervised classification.
The banalization algorithm is also known as If the feature vectors are given with output
the threshold algorithm. The purpose of this label to classifier than it is called supervised
algorithm is to find the suitable threshold classifier. K Nearest Nabors (KNN),
[12]. By use of this threshold value the Artificial Neural Network (ANN) and
targeted image is divided into two parts Support Vector Machine (SVM) are the
foreground and background image. examples of supervised learning approach
Foreground image is represent by white and used in many medical image classification.
background is represent by black color. This In other hand if feature vector is given to the
Mehran University Research Journal of Engineering and Technology
Dated: 28 Apr 2018
classifier without output label then such (SVM) and Convolutional Neural Network
classifier is called unsupervised classifier. K (CNN) are used for classification of normal
means, mixture models, hierarchical and abnormal area in early diagnoses. In
clustering, Principal Component Analysis, future researchers will try to build Image
Singular Value Decomposition are the Processing Techniques on X-ray, PET
example of unsupervised learning methods images for more accuracy. For the
[26]. classification KNN or fuzzy, logic k/c-mean
and clustering may be used. Furthermore,
5.2 MULTI-LABEL contrast between X-ray and CT scan images
CLASSFICATION. for better result for early lung cancer
Multi-label classification in concerned with diagnose.
the classification about information for
different class labels. Multi-label
classification (MLC) is a major research area References
in the machine learning community and
finds application in several domains [1] P. Joon, A. Jatain, and S. B. Bhaskar,
[27].Multi-label classification problem by “Lung Cancer Detection Using Image
adopting a group-sparsity regularize to Processing Techniques,” International
select common subsets of relevant features Journal of Engineering Science, vol.
from different cellular compartments. In 6497, 2017.
addition, we also add a cell structural [2] B. Malik, J. P. Singh, V. B. P. Singh, and
correlation regularized Laplacian term, P. Naresh, “Lung Cancer Detection at
which utilizes the prior biological structural Initial Stage by Using Image Processing
information to capture the intrinsic and Classification Techniques,” Lung
dependency among different cellular Cancer, vol. 3, no. 11, 2016.
compartments[14]. [3] D. Sharma, and G. Jindal, "Identifying
lung cancer using image processing
6.1 CONCLUSION AND FUTURE techniques." pp. 115-120.
[4] M. S. Al-tarawneh, “Lung cancer
SCOPE
detection using image processing
An effort to diagnose the cancer in the lungs
techniques,” Leonardo Electronic
using digital image processing techniques.
Journal of Practices and Technologies,
Initially the CT scan/MRI is captured the
vol. 11, no. 21, pp. 147-58, 2012.
image and processed then cancer or tumor
[5] J. Rani, “Noise Removal In Medical
region is detected and diagnosis from the
Images Using Filters,” International
original image. In pre-processing phase a Journal of Engineering Research &
number of techniques are being used for Technology (IJERT), vol. 2, no. 3, pp.
filter the distortion of the images for 1013-1016, 21 March 2013, 2013.
example Weiner Filter, Fast Fourier [6] M. V. A. Gajdhane, and L. Deshpande,
Transform etc. After pre-processing, image “Detection of Lung Cancer Stages on CT
segmentation phase should be done through scan Images by Using Various Image
Marker-Controlled Watershed and Processing Techniques,” IOSR Journal of
Watershed Segmentation. The next phase is Computer Engineering (IOSR-JCE) e-
features extraction is done by Area, ISSN, pp. 2278-0661, 2014.
Perimeter, and Eccentricity etc. These [7] B. G. Patil, and S. N. Jain, “Cancer cells
features help to detect the lung cancer. In detection using digital image processing
classification, Support Vector Machine methods,” International Journal of
Mehran University Research Journal of Engineering and Technology
Dated: 28 Apr 2018