Diagnosis of Skin Cancer Melanoma using Machine Learning

Ms. Gaana M, Dr Shweta Gupta, Dr Narayana Swamy Ramaiah

JAIN Deemed to be University, Bangalore, Karnataka, India
[email protected]

Abstract

Currently skin cancer is being detected by doctors by manually checking the pattern and area of affected area. Since this
method is time consuming and is prone to human errors, research is being conducted to detect skin cancer automatically.
Image Processing technique are extensively being used for the purpose. However due to limitation of Image Processing
where in certain parameters have to be sent manually, this approach for automated detection turns out practically to be
inefficient thus it would be apt to train the machine to identify the cancer areas without using complex algorithms, which is
turn increases the accuracy and efficiency of the system/algorithms.

Keywords: Machine Learning, Image Processing, Skin Cancer, Melanoma

1. INTRODUCTION

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. The deadliest types in cancers are the Skin Cancer, since it would spread all over our
body and if this not discovered and treated in the early stage. We have 3 types of skin cancers namely Basal cell carcinoma,
Squamous cell carcinoma and Melanoma. But melanoma is deadliest among all skin cancers. The cases of melanoma skin
cancer are increasing day by day. Early detection of Melanoma skin cancer must be cured in initial stage for the patient to
get the chances of high recovery. They are seen on the skin as moles or marks. According to the skin structure it has two
different layers namely epidermis which is considered to be the outer most layer and dermis is the inner layer. The lower part
epidermis has Melanocytes. They contain pigmented cell that is found in epidermis that will produce melanin, this pigment
will give colour to human skin. As the skin is thereby exposed to sun, they produce more and more pigment, hence Causes
skin to be, darkened in colour which causes skin cancer melanoma.

The risk factor of Melanoma is a fairy skin tone, sun burnings, and past genetic factors, prone to less immunity
system and also exposure to ultra violet light rays excessively. If Melanoma is not detected in its early stages it might grow
and also spread along the first layer of skin that is epidermis and at last, it comes in contact with Lymph vessel and the
finally blood. The characteristics are shape, colour and size which are some important parameters for recognizing the skin
cancer and they appear as moles with irregular border, shapes, change in colour, and diameter will be greater than 6mm. The
various non-invasive techniques have been proposed in such a way to analyse detection cancer and separating it to be benign
in nature or melanoma.

2. LITERATURE REVIEW

Dr. Abbas Hannon, Alasadi Baidaa M.ALsafy (2015) says that in order to diagnose Melanoma in early
stages, the image processing technique is used. Here the following processes must be carried out namely, the
initial process to detect the pigment skin lesion is benign or malignant, and the later to recognize malignant
melanoma skin type. The process is further carried out by lesion segmentation and feature extraction. Diagnosis
is carried out to differentiate between benign and malignant melanoma. Hence it can give a different type of
malignant melanoma such as superficial, nodular, lentigo and acral melanoma. [2]

Electronic copy available at: https://ssrn.com/abstract=3358134

 According to Sanjana M and Dr. V. Hanuman Kumar, this paper concentrates on identification of skin
cancer, caused by one of the above-mentioned conditions. The images are processed using combinations of
machine learning and image processing to detect the stage of cancer. Images of the affected area are captured
with the help of derma scope. Several algorithms have been proposed to detect skin cancer but most of the
inputs are fed manually. The main objective of this project is to develop a machine learning algorithm which
requires minimal intervention of human. [3]

Here we have different methods for detection of skin cancer. The detection of melanoma requires various
steps to be followed like pre-processing, the segmentation, classification and feature extraction. This paper
emphasizes on the different methods such as Hybrid Artificial Neural Network, Genetic Algorithm, Neuro
Fuzzy System, and, Hybrid genetic algorithm and Artificial Neural Network, SVM, CNN, ABCD rule and
unsupervised algorithm – K means algorithm etc. However, among all the analysed algorithms, the SVM
algorithm method holds good for detection of cancer because it has least disadvantages. From all the existing
systems we conclude that the neural network technique is best. This information is provided by Nikitha Kaut,
Aayush shah, Shali Vira Harmit Sampat (2018) [4]

Hao Chang who is from Department of Genetics, Yale University School of Medicine says that,
Dermoscopy is one of the most important techniques to exam skin lesions and can capture high resolution
images of the skin escaping the interruption of surface reflections. Especially well-trained clinicians use this
high-resolution imaging to evaluate the possibility of melanoma at the very beginning and can obtain a
diagnostic accuracy as high as ~80%. However, there are not enough experienced dermatologists all over the
world. In order to solve this problem, there has been effort to create machine-driven image analysis software to
classify different skin associated diseases using Dermoscopy images in the academic research community. [5]

Munya A. Arasi ,El-Sayed A El-Dahshan,El-Sayed M. El-Horbaty andAbdelBadeeh M. Salem (2016) have

focused on an overview of recent states of the art in Computer-aided detection/diagnosis (CAD) systems in
identifying and diagnosing malignant melanoma of Dermoscopy images and describe its steps starting with
image acquisition, preprocessing and finishing with malignant melanoma classification of Dermoscopic images.
The comparative study shows that the most common methods for features extraction are the Discrete Wavelet
Transform (DWT) and the method which combines both texture and color features resulting in output of very
high accuracy. The methods for the classification are Nearest Neighbor, Artificial Neural Networks, and Support
Vector Machines are very well. [6]

In this paper authors named Adria Romero Lopez, Xavier Giro-i-Nieto, Jack BurdickandOge Marques
(2015) have stated that the skin lesions classification, mainly in skin cancer melanoma detection, presents a
deep-learning method-based approach in order to solve problems of separating dermoscopic images containing a
skin’s lesion as benign and malignant. This proposed solution is built around VGG Net convolution neural
network architecture and thus it transfer’s learning paradigm. Here we are interested in classification, i.e. to

Electronic copy available at: https://ssrn.com/abstract=3358134

determine a dermoscopic image containing a skin lesion to be a melanoma that is dangerous or benign which is
non-dangerous. [7]

AbrhamDebasuMengistu and DagnachewMelesewAlemayehu (2015) emphasis that 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. [8]

According to this paper, the authors have proposed a framework that consolidates ongoing advancements in
profound learning with set up Artificial Intelligence approaches, making groups of strategies that are equipped
for dividing skin injuries, just as breaking down the identified zone and encompassing tissue for melanoma
recognition. The framework is assessed utilizing the biggest openly accessible benchmark dataset of
dermoscopic pictures, containing 900 preparing and 379 testing pictures. New best in class execution levels are
illustrated, prompting an improvement in the region under beneficiary working trademark bend .More late work
has started to analyse the adequacy of the cutting edge profound learning ways to deal with picture
acknowledgment inside the dermatology and dermoscopy application area .Representations gained from the
characteristic photograph space were utilized, related to unsupervised and hand-coded highlights, to accomplish
best in class execution in an information of more than 2,000 dermoscopy pictures . In any case, the work was
restricted to sore pictures that had been physically pre-divided for example pictures were at that point edited
around the injury. [9]

Sigurdursigurdsson, PeterAlshede Philipsen along with Lars Kai Hansen Jan Larsen plus Monika
Gniadecka and Hans Christian Wolf says that usage of a very robotized counterfeit neural system structure,
factual assessment of neural system perception and an answer for a significant medicinal application. An
element extraction plan was suggested to diminish measurement of the Raman spectra and to smother
foundation commotion starting with the skin’s fluorescence. This proposed neural system’s classifier
consequently keeps away from over-fitting by utilizing versatile regularizations and anomaly detecting. Hence
the neural system structure is circulated for non-business use. Hence forth at last the grouping principles are
extricated from the neural systems with affectability maps. [10]

3. MATERIALS REQUIRED.
▪ Dermoscopic images form the data base [PH2D]
• Here the images of skin cancer are collected from online database where the exact result of the
desired stage is listed.
• As a result, once the data sheet is collected, we accurately check the results.

Electronic copy available at: https://ssrn.com/abstract=3358134

 ▪ MATLAB software Version used: MATLAB R2011b
• MATLAB is worldview nautical processing and exclusively programmed language developed
by Math Works.
• It allows grid controller, plots capacities with information, usage in calculations, making the
UIs with interfacing projects written with the different languages like C, C++, C#, Java,
Fortran and Python.
• In fact, that MATLAB has proposed principally for processing, a discretionary tool kits takes
help of the MuPAD emblematic motors, enables accessing to representative register capacity.
• Extra bundle the Simulink contains graphical multi-spaced reproduction and also model based
plan in dynamic with inserted systems.They have inbuilt capacities which help to perform
confounded operations in single statement.
• Simulink which was developed by Math Works, is a information stream programming
language tool in demonstrating, mimicking and also breaks down multi-area dynamic systems.
• Essential interfacing is graphical outlining instrumentation and adjusting arrangements of
square library. Subsequently they offer a tight integration along the remainder in the
MATLAB conditioning and it either drives MATLAB if not the scripted from it.
▪ Python software Version used: 2.7

4. METHODOLOGY

4.1 Image processing

• Image processing is a technique for changing over a picture into advanced structure.
• In this type processing which the information would be a picture example, a photo and a video
edge utilizing a webcam.
• The aim of processing is a picture and a lot trademark with other parameters to be identified in the
picture.
• Preparing strategies includes the picture a 2D signal and also applying the standardized sign
handling systems.
• An advanced picture holding includes adjusting of computerized information for improving the
picture quality with the guide of PC.
• The handling helps in augmenting the quality, sharpness of picture and details of highlights of
enthusiasm towards extraction of data and further examination.
• The motivation behind processing is that is to watch the items that are not visible, to make better
pictures picture honing, picture.

Steps in Image Processing

▪ Image Acquisition: Once the image is obtained, the various methods of processing are be applied
to the image. In other words, Image acquisition can also be defined as action of reading the image
from some source.

Electronic copy available at: https://ssrn.com/abstract=3358134

▪ Image Pre-processing: This process involves selection of appropriate color model for image to be
processed. The primary aim of Image pre-processing is usually to improve the picture data that
suppresses undesired distortions and there by simply enhances some image characteristics that are
very important to even more processing.
▪ Image segmentation: Image segmentation can be defined being a process of partitioning the
digital image into numerous segments by means of sets involving numerous pixels. The primary
purpose of segmentation is in order to simplify and change the particular representation in the
image in to something that will be even more meaningful. It is usually helping to locate objects
and boundaries like edges, curves and lines etc. A mask is created in such a way that the
background and the affected area are same and color as a result the machine may treat the
background area also to be the skin cancer area as a result to overcome this problem, we create
masking. This process must be carried out carefully.
▪ Dilation: It is a method of morphological operation where in it thickens the object in any binary
image. Levels and direction of this are controlled by its size, shape of the structure’s element. In
MATLAB, dilation is performed by function imerode. It contains the parameters such as, image
and a structuring element.
▪ Erosion: Erosion is another part of morphological operation where its effect is to shrink and thin,
reduce the size of object in the image. The particular amount and extent associated with this
thinning is handled by the shape plus size in the structuring aspect. In MATLAB, morphological
erosion is done by function using imerode. It has parameters such as image and the structuring
element.

4.2 Machine Learning

In machine learning there are two different modes namely training mode and test mode.

Part A: Training mode

Fig.1 Training mode

Electronic copy available at: https://ssrn.com/abstract=3358134

 Part B: Test mode

Fig.2 Test mode

4.3 Histogram.

A histogram is an exhibit of genuine information that uses lines to show the repeat of data
things in dynamic numerical between times of identical size. A Histogram is a graphical introduction of
data using bars of different statures. It is used to shorten discrete or reliable data that are assessed on a
break scale. Right when the components are unending, there are no openings between the bars in any
case, in discrete case gaps should be left between the bars. From the histogram we calculate the mean
value (M). Now using mean value, we need to calculate the suitable equation for displaying the output.
On trying different equation, we can conclude that cubic regression holds good for displaying the
result.

Fig.3 Histogram Curve

5. RESULTS

Stage 1: Benign

Fig.4 Benign

Electronic copy available at: https://ssrn.com/abstract=3358134

 Stage 2: To be Melanoma

Fig. 5 To be Melanoma

Stage 3: Melanoma

Fig 5.3 Melanoma

6. CONCLUSION
In this paper we have used Image acquisition, Pre-processing, Segmentation, Noise removal and Feature extraction.
For the first time we have used Supervised Machine Learning using Cubic Regression. In this method we trained the
machine in such a way that it automatically displays the stage of skin cancer to be Benign, to be Melanoma and Melanoma.

REFERENCES

[1] A. Melerzanov and D. Gavrilov, "MELANOMA DIAGNOSIS USING CONVOLUTIONAL NEURAL NETWORKS OF DEEP
LEARNING", Vrach, vol. 29, no. 6, 2018. Available: 10.29296/25877305-2018-06-06.
[2] A. Alasadi and B. M.ALsafy, "Early Detection and Classification of Melanoma Skin Cancer", International Journal of Information
Technology and Computer Science, vol. 7, no. 12, pp. 67-74, 2015. Available: 10.5815/ijitcs.2015.12.08
[3] dr v hanuman kumar, "International Journal of Research in Advent Technology", Ijrat.org, 2018. [Online]. Available:
http://www.ijrat.org/downloads/Vol-6/dec-2018/Paper%20ID-611201879.pdf.
[4]N. Aayush Shah,and S. HarmitSampat, "A Study on Different Techniques for Skin Cancer Detection", 2016.
[5] Skin cancer reorganization and classification with deep neural network ,Hao Chang, Department of Genetics, Yale University School of
Medicine
[6] El-Sayed A. El-Dahshan, "Malignant Melanoma Detection Based on Machine Learning Techniques", 2016.
[7] Jack Burdick, Oge Marques, "SKIN LESION CLASSIFICATION FROM DERMOSCOPIC IMAGES USING DEEP LEARNING
TECHNIQUES”
[8] Dagnachew Melesew Alemayehu [email protected] Bahir Dar University, "Computer Vision for Skin Cancer Diagnosis and
Recognition using RBF and SOM."
[9] N. A. Halpern, and J. B. Helba, "Deep learning ensembles for melanoma recognition in dermoscopy images". .", Ijmlc.org, 2019.
[Online]. Available: http://www.ijmlc.org/vol8/664-Q0010.pdf.
[10] M. Hans ChistianWulf and b. Peter AlshedePhilipsen, "Detection od skin cancer by classification of Raman spectra."

Electronic copy available at: https://ssrn.com/abstract=3358134