cancer rates have been increasing for the past few decades. The risk factor is
the direct exposure of skin lesions to UV radiation which causes various skin
diseases. Skin cancers are most common disease and are deadly to the human.
Early detection of skin cancer can be cured. With the latest technologies,
early detection is possible. One of such technique is artificial intelligence.
The dermoscopy image is given as input and it is processed for noise filtering
and image enhancement. Then the image is segmented using thresholding. A cancerous
skin has certain features and such features are extracted using feature
extraction. These features are given as input to the neural network. The Neural
network is used to classify whether it is cancerous or non-cancerous.
Keywords-skin cancer,artificial intelligence,neural
network,segmentation (key words)
affects the skin is called skin cancer. Skin cancer is of two types malignant
or benign form. Benign Melanoma is the appearance of moles on the skin it is
not a deadly one. Malignant melanoma is the appearance bleeding sores. It is
the deadliest form of all skin cancers. It arises from cancerous growth in pigmented
skin lesion. If it is diagnosed at the right time, this disease is curable. But
diagnosis is difficult. It needs sampling and laboratory tests. Through
lymphatic system or blood melanoma can spread to all parts of the body. So
automatic detection will be useful at these cases. Basically skin disease
diagnosis depends on the different characteristics like color, shape, texture
etc. there are no accepted treatment for skin diseases Different physicians
will treat differently for same symptoms. Key factor in skin diseases treatment
is early detection further treatment reliable on the early detection. In this
paper, Proposed system is used for the diagnosis multiple skin disease using artificial
intelligence and neural network.
This paper is organized as
follows: Section I gives the introduction about Skin cancer and features of
skin cancers. Section II describes the Automatic Skin cancer Detection system
and various steps involved in the system. Section III gives the explanation of
various algorithms used yet. Section IV describes about the proposed system.
Section V concludes the paper followed by references.
III. Automatic skin cancer detection
The first stage of any system is the acquisition of input
image after the input image is obtained various process can be done on it to
obtain the desired output here we do for image pre-processing and segmentation.
CMOS camera is used as the medium for the acquisition of input image.
is the process of removing noise
from an image. Noise can be random or white noise with no coherence, or
coherent noise introduced by the device’s mechanism or processing algorithms. By doing this process we can obtain a good quality of
image for the further segmentation.
This approach is a
displaying system that takes in a practical mapping from an information picture
to a yield picture. The information picture is the first picture, and the yield
picture is a division cover. This empowers the system to show useful residuals,
and additionally to supply higher determination data to the yield layers, so as
to enhance execution of the system in contrast with systems without the skip
associations. The exactness of the division procedure extraordinarily
influences ensuing component extraction and order. Factors Concerning the Segmentation
Various factors that affect the
segmentation of skin cancer images are as follows:
• The skin lesions have complex
structure, large variations in size as well as complex colours in the skin.
• The lesion is contrast to the
• The borders of lesions are not
always well defined.
• The influence of small
structures, hairs, bubbles, light reflection, and other artifacts.
• The influence of the skin
lesions in the surrounding regions.
The highlights which have been utilized to
portray the skin sore pictures are depicted. In this work, we utilize shading,
surface, and shading histogram highlights to speak to injury zones. The purpose
of picking these sorts of highlights is a result of the way that shading and
surface are the main properties commanding in the sore region. Feature
extraction is the critical device which can be utilized to dissect and
investigate the picture properly.
They include extraction depends on the ABCD manage of
dermatoscopy. The ABCD remains for Asymmetry, Border structure, Color variety
and Diameter of sore. It characterizes the reason for the conclusion of a malady.
Injury grouping is the last advance. So
as to arrange a picture grouping strategies like SVM method is used:
USING SUPPORT VECTOR MACHINE:
Vector Machines depend on the idea of choice planes. A choice plane is
otherwise called a hyper plane that isolates between arrangements of items
having distinctive class enrollments.
isolating line characterizes a limit on the correct side of which all s are
GREEN and to one side of which all items are RED. That is all focused on one
side of the hyper plane are named yes, while the others are delegated no.
algorithm of SVM classifier is given as
1. Definition of
Classification Classes –
upon the goal and the qualities of the picture information, the order classes
ought to be unmistakably characterized.
2. Selection of Features –
Highlights to separate between the classes
ought to be set up utilizing multispectral as well as multi-transient
attributes, surfaces and so on.
3. Sampling of Training Data –
Preparing information ought to be inspect
keeping in mind the end goal to decide proper choice tenets.
4. Estimation of Universal
Different arrangement procedures will be
contrasted and the preparation information, so that a suitable choice lead is
chosen for ensuing grouping.
5. Classification –
In light of the choice administer, all pixels are ordered in
a solitary class. There are two techniques for pixel by pixel arrangement and
per – field grouping, regarding divided zones.
In fuzzy logic algorithm, a combination of both ABCD
(Asymmetry, Boarder factor, Color factor, Diameter) rules and Wavelet
coefficients has been used to improve the image feature classification accuracy
In this, the percentage of red, blue,
green is calculated using,
Red% = R÷ (B+G) ×100
Blue% = B÷(R+G) ×100
Green% = G÷ (B+R) ×100
C1-GREEN is calculated here in
order to determine if R/B/G is
dominant over the other, Wavelet transform, Deconstruction, Reconstruction: The wavelet is repeated as,
W (j) = W (j+1) + U (j+1)
Fuzzy interference decision system
will give us quantitative information about ABCD factors which is used with
fuzzy interference system further. Accuracy is 60% only.
KNN remains for k-closest neighbour calculation; it is one of
the easiest yet generally utilized machines learning calculation. A protest is
ordered by the distance from its neighbours with the question being doled out
to the class most basic among its k separate closest neighbours. On the off
chance that k = 1, the calculation just turns out to be closest neighbor calculation,
what’s more the protest is characterized to the class of its closest neighbour.
The downside of k
closest neighbours classifier is, it is influenced by the quantities of features.
The result might be because of the solver whose undertaking in little component
space is harder than in bigger ones. Truth be told, as the dimensionality
expands then the arrangement issue turns out to be all the more directly
detachable, which tends to facilitate the assignment of finding a legitimate
isolating hyper plane. Hence, the preparation time will be longer when compared
C.Artificial Neural Network
An Artificial Neural
Network (ANN) is a data handling that is roused incidentally organic sensory
systems, for example, the mind, process the data. An ANN is arranged for a
particular application, for example, design acknowledgement or information
order, through a learning procedure. A prepared neural system can be thought of
as a “specialist” in the class of data it has been given to break
case of a medical field, error rates of ANN were high when compared to SVM in
which 82.7% test set correctness has been achieved.
SVMs are presently a hotly debated issue in the machine
learning group, making a comparative eagerness at the minute as Artificial
Neural Networks used to do some time recently. Far being, SVMs yet speak to an
effective method for general (nonlinear) grouping, relapse and anomaly discovery
with a natural model portrayal. Bolster vector machines are an arrangement of
related regulated learning strategies utilized for grouping and relapse. Given
an arrangement of preparing cases, each set apart as having a place with one
of two classifications, a SVM preparing
calculation assembles a model that predicts whether another illustration falls
into one classification or then again the other. So, when compared to all above
methods, SVM is good to go.
In this project we have designed a diagnosis system based
on the techniques of image processing. This work is done on different skin
patterns and tones of images and it is analyzed to obtain the result whether
the person is suffering from skin cancer or not. This system helps in the early
detection and cure of skin cancer .this is cost effective and feasible test
method for the detection of skin cancer. The below mentioned is the block of
the early detection skin cancer analyzer.
A).Colour image to gray scale
As the skin tone of people may differ,
based on their region of living this may affect the efficiency in the output
.so in our project we convert the image to gray scale image to increase the
efficiency of the output.
Restoration is the process of recovering the degraded image from a blurred and
noisy one. The degraded images can be stored in different ways. Such as
imperfection of imaging system, bad focusing, motion and etc are the various
defects which cause image degradation. The corrupted images lead to fault
detection, therefore, to select the most appropriate denoising algorithm it is
essential to know about noises present in an image. The image noises can be
divided into four groups of Gaussian, Salt and Pepper, Poisson and Speckle. The
sample of such noises has been shown below,
a) Image without noise b) Gaussian
noise c) Poison noise d) Salt and Pepper noise e) Speckle noise
Mean filters: It works best
with Gaussian noise and for salt and pepper noise. Although this filter reduces
the noise, blur the image and reduce sharp edges.
– Arithmetic mean filter: It
is the simplest of mean filter. It can uniform the noise and works well with
– Geometric mean filter: It
can preserve the detail information of an image better than the arithmetic mean
– Harmonic mean filter: It
works well with salt noise, and other types of noise such as Gaussian noise,
Work well with pepper noise.
– Contra harmonic mean filter: It can
preserve the edge and remove noise much better than arithmetic mean filter.
Because of preserving the edges
character we use harmonic and contra harmonic filters in this system.
Removing Thick Hairs
Though the and
skin lines such as rashes, moles will be smoothed using restoration filters,
the image may include the hairs. Thick hairs in automated analysis of small
skin lesions are considered to mislead the segmentation process. To remove the
thick hairs in skin cancer images, methods such as mathematical morphology methods,
curvilinear structure detection, and automated software called Dull Razor and
Top Hat transform combined with a bicubic interpolation approach are preferred.
The hair-free images are acquired using these operations.
At the end of pre-processing step of skin cancer
detection system, the resulting images are distinguishable from those initial
D). Image enhancement
equalization the technique of adjusting image intensities for enhancing the
contrast. It is one of the non-linear contrast enhancement technique. Let
f be a given image represented as mr
/ mc matrix of integer pixel intensities ranging from 0 to L ? 1.
the number of possible intensity values. Often it will be 256.
p is the normalized histogram of f with a bin
for each possible intensity.
So , pn = number of pixels with intensity n
total number of pixels
where, n = 0, 1, …,
L ? 1.
Canny edge detection uses multi stage algorithm for detecting
wide range of edges in the image. The general
criteria for edge detection include:
Detection of edge with low error rate, which means that the
detection should accurately catch as many edges as possible
The edge point detected from the operator should accurately
localize on the center of the edge.
A given edge in the image should only be marked once, and where
possible, image noise should not create false edges.
The Process of Canny edge detection
algorithm can be broken down to 5 different steps:
In order to remove
the noise apply Gaussian filter to smooth the image
The intensity gradients of the image has
suppression to get rid of spurious response to edge detection
Apply double threshold
to determine potential edges
Track edge by hysteresis: Finalize the detection of edges by
suppressing all the other edges that are weak and not connected to strong
It is the input image ,
It is the gray scale converted and enhanced
It is the canny edge detected image
extraction is done using the properties called ABCDE in automated diagnosis of
skin cancer. ABCDE represents Asymmetry, Border, Colour variation, Diameter and
Asymmetry: Asymmetric nature of
melanoma is property in which the imaginary line passing through middle of
lesion, either up or down or side to side gives two unequal or two
non-symmetric parts. Degree of asymmetry can be calculated by using asymmetric
Index which is calculated by using the formula,
AI = (?A/A) × 100, where A is
the total area of the image and ?A is the difference in area between total
image and lesion area.
Border irregularity: The
border or edge of the skin cancer affected area will be usually blurred or
ragged or irregular or notched. Border irregularity is usually calculated by
compact index in medical image processing. Compact index is used to estimate
unanimous 2D objects. The measure is sensitive to noise along the boundary.
Compact index is calculated using the formula,
where Pl is Perimeter of the
Lesion and Al is area of the Lesion.
Colour variation: Emergence
in colour variation can be detected if lesion is melanoma. The colours can be
variations in black, brown and red depending on the production of melanin
pigment in the affected area. Colour variation can be detected statically and
by plotting histograms of the segmented image. The intensity variation is high
if there are colour variations.
Diameter: Skin cancer
(melanoma) usually have diameter more than 6mm. Since diameter is irregular, it
is calculated by drawing from edge pixels to pixels in the midpoint and
G). Classifier based on neural networks
A neural network consists of units
(neurons), arranged in layers, which convert an input vector into some
output. Each unit takes an input, applies a (often nonlinear) function to
it and then passes the output on to the next layer. Generally the
networks are defined to be feed-forward: a unit feeds its output to all the
units on the next layer, but there is no feedback to the previous layer.
Weightings are applied to the signals passing from one unit to another, and it
is these weightings which are tuned in the training phase to adapt a neural
network to the particular problem at hand. This is the learning phase.Neural
networks have found application in a wide variety of problems. These
range from function representation to pattern recognition, which is what we
will be consider here.
Thus as the earlier mentioned the
SVM classifier is used here to classify whether it is a cancerous or non-cancerous
i.e., benign or malignant cancer. The layered architecture of neural network is
being used here for the classification purpose. The efficiency and the accuracy
are expected up to 98 %. But the efficiency may vary according to the type of
segmentation and classifiers used in it.
based early detection of skin cancer analyzer system is being proposed. It has
been found to be a better diagnosis method than the artificial and k-nearest
neighbor methods. This methodology uses image processing and support vector
machine for classification of malignant melanoma from other skin diseases.
Dermoscopic image were collected and processed by various image processing
techniques. The cancerous region is separated from the healthy skin by the
method of segmentation. Based the features the images are classified as
cancerous or non- cancerous. It has got good accuracy and efficiency of 98%
also. By further varying the image processing techniques and classifiers
accuracy and efficiency can be improved for this system.
We would like to thank our guide and professors
Electronics and communication Engineering Department, M.Kumarasamy College of
Engineering management for their guidance and support and facilities extended