Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM

Abrham Debasu Mengistu & Dagnachew Melesew Alemayehu
International Journal of Image Processing (IJIP), Volume (9) : Issue (6) : 2015 311
Computer Vision for Skin Cancer Diagnosis and Recognition
using RBF and SOM
Abrham Debasu Mengistu abrhamd@bdu.edu.et
Bahir Dar University, Bahir Dar institute of technology
Faculty of computing
Bahir Dar, Ethiopia
Dagnachew Melesew Alemayehu dagnachewm@bdu.edu.et
Bahir Dar University, Bahir Dar institute of technology
Faculty of computing
Bahir Dar, Ethiopia
Abstract
Human skin is the largest organ in our body which provides protection against heat, light,
infections and injury. It also stores water, fat, and vitamin. Cancer is the leading cause of death in
economically developed countries and the second leading cause of death in developing
countries. Skin cancer is the most commonly diagnosed type of cancer among men and women.
Exposure to UV rays, modernize diets, smoking, alcohol and nicotine are the main cause. Cancer
is increasingly recognized as a critical public health problem in Ethiopia. There are three type of
skin cancer and they are recognized based on their own properties. In view of this, a digital
image processing technique is proposed to recognize and predict the different types of skin
cancers using digital image processing techniques. Sample skin cancer image were taken from
American cancer society research center and DERMOFIT which are popular and widely focuses
on skin cancer research. The classification system was supervised corresponding to the
predefined classes of the type of skin cancer. Combining Self organizing map (SOM) and radial
basis function (RBF) for recognition and diagnosis of skin cancer is by far better than KNN, Naïve
Bayes and ANN classifier. It was also showed that the discrimination power of morphology and
color features was better than texture features but when morphology, texture and color features
were used together the classification accuracy was increased. The best classification accuracy
(88%, 96.15% and 95.45% for Basal cell carcinoma, Melanoma and Squamous cell carcinoma
respectively) were obtained using combining SOM and RBF. The overall classification accuracy
was 93.15%.
Keywords: SOM, RBF, KNN, Digital Image Processing, Dermofit.
1. INTRODUCTION
Human Skin Cancer is a disease that appears on the outer layers of the skin which are caused
when the skin cells are dead or damaged due to over exposure to Sun’s ultraviolet radiation. But
Skin cancer can also occur on areas of one’s skin not ordinarily exposed to sunlight. The human
skin is the largest organ in our body which provides protection against heat, light, infections and
injury. It also stores water, fat, and vitamin D. [1] The Human skin is composed of two major
layers called epidermis and dermis. The top or the outer layer of the skin which is called the
epidermis composed of three types of cells flat and scaly cells on the surface called SQUAMOUS
cells, round cells called BASAL cells and MELANOCYTES, cells that provide skin its color and
protect against skin damage. The inner layer of the skin known as the dermis is the layer that
contains the nerves, blood vessels, and sweat glands. There are three type of skin cancer
Melanoma, Basal cell carcinoma and Squamous cell carcinoma. Skin cancer is diagnosed by
physical examination and biopsy. Biopsy is a quick and simple procedure where part or all of the
spot is removed and sent to a laboratory. It may be done by doctor or you can be referred to a

dermatologist or surgeon. Results may take about a week to be ready [1]. Dermatology imaging
research believes that diagnosis of skin cancer can be automated based on a certain physical
feature and color information which is the characteristics of different category of skin cancer.
FIGURE 1: Layers of Skin.
2. LITERATURE REVIEW
Regarding with the benefits of early detection of human skin cancer, several dermatologists,
medical professionals, medical industries, clinicians, computer scientists, academic researchers
and technical experts have dedicated time and efforts to improve the early detection of
human skin cancers. And some researchers have been done on the application on neural
classifiers to skin injury classification purposes on this research paper presents a computer
vision approach based on image processing algorithms and supervised learning techniques to
help detecting and classifying wound tissue types which play an important role in wound
diagnosis. The system proposed involves the use of the k-means clustering algorithm for image
segmentation and a standard multilayer perceptron neural network to classify effectively each
segmented region as the appropriate tissue type. [11]. Common classification method like
statistical and rule based ones were applied in the researches to describe the diagnostic
performance of Solar Scan, an automated instrument for the diagnosis of primary melanoma
Images from a data set of 2430 lesions (382 were melanomas) were divided into a training set
and an independent test set Solar Scan is a robust diagnostic instrument for pigmented or
partially pigmented melanocytic lesions of the skin. [11].
More advanced techniques such as Neural Network were presented in the research [12]
The aim of the study was to provide mathematical descriptors for the border of pigmented skin
lesion images and to assess their efficacy for distinction among different lesion groups. New
descriptors such as lesion slope and lesion slope regularity are introduced and mathematically
defined descriptors was tested on a data set of 510 pigmented skin lesions, composed by 85
melanomas and 425 nevi, by employing statistical methods for discrimination between the two
populations.
K-nearest neighborhood as another classification method was employed in the research of [12]
on this research paper presents an algorithm for classification of non melanoma skin lesions
based on a novel hierarchical K- Nearest Neighbors (KNN) classifier. The KNN classifier here,
skin lesions are characterized by their color and texture. Finally, towards identification of human
skin cancer uses the following common steps is image acquisitions, preprocessing,
Segmentation, feature extraction, classification and finally the result will display to the user. [13]
In the recent years computational vision based diagnostic systems for dermatology have
demonstrated significant progress. In this work, they review these systems by firstly presenting
the visual features used for skin lesion classification and methods for defining them. Then they
describe how to extract these features through digital image processing methods, i.e.,
segmentation, registration, border detection, color and texture processing) [14]. However, these
imaging technologies are still expensive and may require specialized training to read the resulting
images. Dermoscopy is the methodology for the exanimation of skin injuries based on imaging.
This method provides a good and detailed view of the injuries. The imaging equipment used for
taking the images is called Dermatoscope. It is handheld equipments which is compact and easy

to use. An oil film is placed between the lens of detematoscope and skin injuries. Main purpose of
placing oil film is to obtain the magnified view of skin tissues [15].
The conventional diagnosing method for skin cancer is biopsy. It is a painful and time consuming
technique. By incorporating Artificial intelligence and Digital Image Processing for skin cancer
detection, it is possible to do the diagnosis without any physical contact with the skin. This can be
implemented in a computer with the help of some software. Skin cancer detection system
implemented using computer and software is known as Computer Aided Detection. The
detection system is mainly based on Artificial intelligence and Digital Image Processing. Artificial
intelligence has proven to be very efficient in decision making and pattern recognition
applications. In this paper, the ANN Classifier is implemented in MATLAB software for skin
cancer detection [16]. First stage in the skin cancer detection system is the input image.
Dermoscopic image in digital format is given as input to the system. Dermoscopic image in digital
format is given as input to the system. Next stage is the noise removal. The image contains hairs
and other noises. These noises cause errors in classification. The noises are removed by filtering.
Filtering method implemented here is the Median Filtering [16].
3. PROBLEM STATEMENT
On the previous researches there is a scope for the design of classifier to detect the type of
cancer this provide a better and more reliable results for the patients, so that more patients can
be diagnosed and cured. In line with this, human skin cancer identification is very useful in
encouraging good quality in skin cancer diagnosis. There is a need for automated in recognition
of human skin cancer systems so that the abuses during diagnosis and treatment can be
minimized. Therefore, this thesis work will initiate a model for human skin cancer recognition
which is consistent, efficient and cost effective by exploring the technology of image processing
techniques. The ultimate goal is to ease the doctor's role in the recognition of skin cancer
mentioned above by providing better and more reliable results, so that more patients can be
diagnosed. The work on classifier design to detect the type of cancer will be taken in future [31].
Skin cancer is diagnosed by physical examination and biopsy. In case of physical examination
the doctors will try to see the physical properties of skin cancers. When we see biopsy it is the
procedure that the dermatologist takes some part or all of the spot and sent to a laboratory. It
may be done by doctor or you can be referred to a dermatologist or surgeon. Results may take
about a week to be ready [1]. To this end this study answers the following research question:
What appropriate image processing techniques used for human skin cancer recognition?
To what extent recognition effectiveness is registered for the human skin cancer?
What are the features that distinguish the three type of skin cancer?
What are the common features that the three type of skin cancer shares?
How to develop an automatic skin cancer recognition system based on image processing
techniques?
4. DESIGN OF HUMAN SKIN CANCER RECOGNITION
The task of recognition occurs in wide range of human activity. The problem of recognition is
concerned with the construction of a procedure that will be applied to differentiate items, in which
each new item must be assigned to one of a set of predefined classes on the basis of observed
attributes or features.
Accordingly, image analysis or computer vision is used in the recognition of human skin cancer to
predefined classes. The predefined classes are the feature or attributes are computed from skin
cancer images. These observed features of skin cancer were used to decide the class or the type
of skin cancer. Hence, in this research the main interest is to differentiate the type of skin cancer
varieties by using image analysis technique this is because of in order to maximize the curability
of the disease if we identify the type of skin cancer where it belongs to it is very simple to cure the
disease otherwise it is difficult.

FIGURE 2: Skin Cancer Recognition Process Model.
As shown in the above figure 2, classification of human skin cancer involves the following
activities:
Image acquisition of human skin cancer, an image processing techniques is applied on the
acquired image to enhance the quality of image so as to reduce noises, appropriate feature are
extracted from the enhanced image by using image analysis techniques that is used to classify
dermoscopy images of skin cancer, the classification model of training and testing data of
dermoscopy images of skin cancer will be developed, finally suitable pattern classifiers are
selected to classify dermoscopy images of skin cancer to the predefined classes of skin cancer.
A. Image Acquisition
Image analysis starts with image acquisition this involves all aspects that have to be addressed in
order to obtain dermoscopy image of human skin cancer the selection of radiation (light) sources
and sensors (such as cameras) has to be considered very carefully. For this study, images have
been taken from the following websites:
https://www.dermquest.com/image-library/
http://www.dermnet.com/images/
B. Image Processing
Image processing is a mechanism that focuses on the manipulation of images in different ways in
order to enhance the image quality. Images are taken as the input and output for image
processing techniques it is the analysis of image to image transformation which is used for the
enhancement of image i.e. to increase the contrast for the input image and also restoration for
geometrical distortion. [10]. Image segmentation is one of the most important tasks in image
processing. It is the process of dividing an image into different homogeneous regions such that
Skin Cancer
Image
Enhanced
Image
Extracted
features
Classifier
Skin Cancer
Recognition
(Type)
Image Processing
Image Analysis

the pixels in each partitioned region possess an identical set of properties or attributes [10]. The
result of segmentation is a number of homogeneous regions, each having a unique label. Image
segmentation is basically used to isolate region of interest from the background noise. For image
processing techniques, we have used Matlab R2013a in which MATLAB is a high performance
language for technical computing. MATLAB (MATRIX LABORATORY) is an interactive system for
matrix based computation designed for scientific and engineering use. It is good for many forms
of numeric computation and visualization. Hence, MATLAB was used for image processing tasks
of Human skin cancer images to enhance the quality of image and to change images to binary for
feature extraction purposes.From the original skin cancer images, the image is filtered in order to
avoid other noises that are formed due to illumination effects as shown figure below.
FIGURE 3: Image Processing.
C. Median Filtering
The Dermoscopic Image in Digital format is subjected to various Digital Image Processing
Techniques. The standard image size is taken as 360x360 pixels [27]. Usually the image
consists of noises in the form of hairs, bubbles etc. These noises cause inaccuracy in
classification. In order to avoid that, images are subjected to various image processing
techniques. One of the key element in image processing is filtering of image pre processing is
done to removes the noise, fine hair and bubbles in the image. For smoothing image from noise,
median filtering is used. Median filtering is a common step in image processing. Median filtering
is used for minimizing the influence of small structures like thin hairs and isolated islands of pixels
like small air bubbles [30].
D. Feature Extraction
Feature extraction is the method by which unique features of skin lesion images are
extracted. This method reduces the complexity in classification problems. The purpose of feature
extraction is to reduce the original data set by measuring certain properties, or features, that
distinguish one input pattern from another. We have the following three groups of features:
GLCM (Texture features of skin cancer): GLCM is a powerful tool for image feature extraction
by mapping the grey level co occurrence probabilities based on spatial relations of pixels in
different angular directions.
Morphological features: Morphology is the geometric property of a given image, in our case it is
the size and shape characteristics of human skin cancer image.

Basalcell Melanoma Squamouscell Basalcell Melanoma Squamouscell Basalcell Melanoma Squamouscell Basalcell Melanoma Squamouscell
Basalcell 18 3 4 Basalcell 15 5 5 Basalcell 14 4 7 Basalcell 22 2 1
Melanoma 3 20 3 Melanoma 6 18 2 Melanoma 8 14 4 Melanoma 1 25 0
Squamouscell 6 2 14 Squamouscell 7 2 13 Squamouscell 7 2 13 Squamouscell 0 1 21
Total 73 Total 73 Total 73 Total 73
correct 52 correct 46 correct 41 correct 68
notcorrect 21 notcorrect 27 notcorrect 32 notcorrect 5
% 71.232877 % 63.013699 % 56.164384 % 93.150685
KNN ANN Naïve SOM
Color features: Color is one of the features of skin cancer, they have different color variation of
each cancer type and color analysis computed by taking the mean value of RGBs (Red, Green
and Blue) components and the mean value of HSIs (Hue, Saturation and Intensity) components.
Therefore, to compute the mean value of each component of these color spaces, we use matlab
2013 to split each component because matlab has built in function to convert to HIS or RGB color
spaces. By using the built in function of matlab we can find RGB, the RGB color image stack is
split to red, green and blue components. Hence, the color features are extracted by computing
the mean values of RGBs and HSIs of Dermoscopy skin cancer images. That is, the mean value
of red, mean value of green, mean value of blue, mean value of hue, mean value of saturation
and mean value of intensity are computed from each component.
5. EXPERIMENTAL RESULTS
We extract 15 features (four morphology, five GLCM and six color features) were identified;
hence, the total input features were fifteen. These features were used to classify different skin
cancer image of human body. In line with this, we have designed experimental scenarios to test
the classification performance by taking the extracted features of cancer images. The
classifications were tested by four different algorithms namely ANN (Artificial Neural Network),
KNN (Nearest Neighbor classification) , Naive Bayes and combining SOM and RBF classifiers in
order to get a more accurate result. In order to train the classifiers, a set of training skin cancer
image was required, and the class label where it belongs to, 235 skin cancer image were taken
from American skin cancer society and DERMOFIT from the predefined three types of skin
cancer i.e. Melanoma, Basal cell carcinoma and Squamous cell carcinoma.
There are two basic phases of pattern classification. They are training and testing phases. In the
training phase, data is repeatedly presented to the classifier, in order to obtain a desired
response. In testing phase, the trained system is applied to data that it has never seen to check
the performance of the classification. Hence, we need to design the classifier by partitioning the
total data set into training and testing data set. From the total data set of each skin cancer type
70% was used to build training and the remaining 30% of the total was used for testing data.
From the total of 235 data sets, 162 were used for training and 73 were used for testing. In
general, a classier has some input features based on the scenario of the designed experiment
and some output features. In this study, there were three output classes, because the predefined
human skin cancer images were three. The total numbers of exemplas or patterns were 235. This
exemplas were normalized with mean 0 and variance 1.
FIGURE 4: Summary result of KNN, ANN, Naïve and SOM classifier using extracted features.
As we have presented in detail in the previous section, the experiments were conducted under
four scenarios by using feature sets of morphology, texture and color separately, and finally
combining the three feature sets. Then, the experiment results were compared the performance
of the Naïve Bayes classifier, KNN, neural network and SOM classification over the three

0
5
10
15
20
25
30
Basal cell Melanoma Squamous cell
SOM
scenarios. The total number of data sets is 235. Out of these, 70% were used for training and the
remaining 30% were used for testing. In general, the overall result showed that morphology and
color features have more discriminating power than texture features and the classification
performance of SOM is by far better than Naïve Bayes , artificial neural network classifier and
KNN.
FIGURE 5: Overall Performance.
FIGURE 6: Confidence Interval of Detection.
6. CONCLUSION AND FUTURE WORK
In the classification problem of skin cancer recognition, morphological, GLCM and color features
were extracted from a skin cancer images taken from three type of skin cancer Basal cell
carcinoma, Melanoma and squamous cell carcinoma by using image analysis techniques. These
selected features were used as input to the classification model. The experiment was conducted
under four scenarios of the features data set such as GLCM, Morphology, Color and combining
the three features. The result of the experimentation showed that the three varieties of Human
skin cancer have been classified more accurately by SOM than using Naïve Bayes , ANN and
KNN classifier. The image analysis for the recognition of the type of skin cancer can be further

investigated. The work can also be seen in depth and researched by the different characteristics
of its physical and chemical in connection to image technology.
In light with this, the following recommendations are made for further research and
improvements.
Identification of skin cancer type by exploring more features, Skin cancer recognition by levels of
injuries using image analysis, implementing skin cancer recognition on mobile to make simplified
for doctors, Computer vision for Recognition of leprosy and skin cancer.
7. REFERENCES
[1] http://www.cancer.org/cancer/skincancer
[2] Cancer in Africa, World Health Organization, International Agency for Research on cancer;
2009.
[3] http://www.skincancer.org/skin-cancer-information/skin-cancer-facts
[4] http://www.cdc.gov/cancer/skin/statistics/
[6] World Health Organization, the global burden of disease: 2008 Update Geneva: World
Health Organization,2008
[7] A Morrone, Skin Cancer in Ethiopia,21st world congress of dermatology, 2007
[9] William K. Pratt: Digital image processing, PIKS Scientific inside, John Wiley, 4th Edition,
2007
[10] John Breneman towards early stages of malignant melanoma detection Using Consumer
Mobile Devices
[11] Luís Filipe Caeiro M argalho Guerra Rosado, Automatic System for Diagnosis of Skin
Lesions Based on Dermoscopic Images
[12] T Y Satheesha,, Dr. D Sathya Narayana, Dr. M N Giriprasad, review on early detection of
melanoma:2012
[13] Ilias Maglogiannis, Dimitrios I. Kosmopoulos, Computational vision systems for the
detection of malignant melanoma
[14] Ioana Dumitrache, Alina Elena Sultana, and Radu Dogaru Automatic Detection of Skin
Melanoma from Images using Natural Computing Approaches
[15] Aswin.R.B, J. Abdul Jaleel, Sibi Salim, Implementation of ANN Classifier using MATLAB for
Skin Cancer Detection,2013
[16] Ioana Dumitrache, Alina Elena Sultana, and Radu Dogaru Automatic Detection of Skin
Melanoma from Images using Natural Computing Approaches
[17] Nilkamal S. Ramteke and Shweta V. Jain ABCD rule based automatic computer-aided skin
cancer detection using MATLAB
[18] Santosh Achakanalli and G. Sadashivappa Skin Cancer Detection and Diagnosis Using
Image Processing and Implementation Using Neural Networks and ABCD Parameters

[19] Mariam A.Sheha and Mai S.Mabrouk Automatic Detection of Melanoma Skin Cancer using
Texture Analysis
[20] Catarina Barata, Margarida Ruela, Mariana Francisco, Teresa Mendonça, and Jorge S.
Marques Detection of Melanomas in Dermoscopy Images Using Texture and Color
Features
[21] Santosh Achakanalli & G. Sadashivappa, Skin Cancer Detection And Diagnosis Using
Image Processing And Implementation Using Neural Networks And ABCD Parameters
[22] Luís Filipe Caeiro M argalho Guerra Rosado, Automatic System for Diagnosis of Skin
Lesions Based on Dermoscopic Images
[23] Mariam A. Sheha , Mai S. Mabrouk , Amr Sharawy, Automatic Detection of Melanoma Skin
Cancer using Texture Analysis
[24] Lin Li, Qizhi Zhang, Yihua Ding, Huabei Jiang, Bruce H Thiersand, Automatic diagnosis of
melanoma using machine learning methods
[25] Sarika Choudhari, Seema Biday, Artificial Neural Network for SkinCancer Detection
[26] Dr. J. Abdul Jaleel, Sibi Salim, Aswin.R.B, Diagnosis and Detection of Skin Cancer Using
Artificial Intelligence
[27] Peyman Sabouri, Hamid GholamHosseini, John Collins, Border Detection of Mela noma
Skin Lesions, 2013
[28] Dr. J. Abdul Jaleel, Sibi Salim, Aswin.R.B, Artificial Neural Network Based Detection of Skin
Cancer,2012
[29] Snehal Salunke, Survey on Skin lesion segmentation and classification,2014
[30] Nandini M.N., M.S.Mallikarjunaswamy, Detection of Melanoma Skin Disease using
Dermoscopy Images
[31] Tinku Acharya and Ajoy K. Ray, Image Processing Principles and Applications, Jhon Wiley,
2005.
[32] William K. Pratt: Digital image processing, PIKS Scientific inside, John Wiley, 4th Edition,
2007.

Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM

More Related Content

What's hot (19)

Similar to Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM (20)

Recently uploaded (20)

Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM