Detection of Melanoma Skin Cancer

Deekshitha R Lavanya V Mounika M K L

Dept. of Computer Science &Engg Dept. of Computer Science &Engg Dept. of Computer Science &Engg
K.S Institute of Technology K.S Institute of Technology K.S Institute of Technology
Bengaluru,India Bengaluru,India Bengaluru,India
manupatideekshu@gmail.com lavanya08.vasanth@gmail.com mounika.marrey@gmail.com

Neha K Mrs.Beena K
Dept. of Computer Science &Engg Assistant Professor
K.S Institute of Technology Dept. of Computer Science &Engg
Bengaluru,India K.S Institute of Technology
nehakunapalli623@gmail.com Bengaluru,India
beenak@ksit.edu.in

Abstract – Dermatology diseases are one of the biggest medical 6. Test case to load the image and identify the test
issues in 21st century due to its highly expensive and complex image is benign or malignant.
diagnosis. Skin cancer are mainly of 2 types – Melanoma and Non 7. Graphical User Interface to use the program
-melanoma. Early detection of this fatal Melanoma skin disease functionality.
increases the curing rate to 90%. In diagnosis of medical images
,vision of computer can play a vital role and is also proved by
many existing system.
III. METHODOLOGY

The system consists of four stages:

I. INTRODUCTION
A. DATASET COLLECTION:
The appropriate dataset was collected from International
SKIN- Jack of all trades meaning it performs various tasks Skin Imaging Collaboration (ISIC) Archive. Goal of the
like protects the body by covering all the muscles and other archive is to serve as image resources for public for research
tissues, maintaining the body optimal temperature in and development, teaching, testing of diagnostic AI
extreme climatic conditions, helps as an excretory organ algorithms.
removing wastes through sweating. Melanocyte if present in
human skin causes Melanoma. Our task is mainly centered
around detection of this lethal type of malignant growth.
Melanoma can be detected by straightforward visual
assessment known as ABCDE examination where A stands
for ASSYMETRIC cancerous mole tend to be irregular in
shape. B stands for BORDER it appears to be ragged and
notched while C stands for COLOR includes uneven shades
of colors. D stands for DIAMETER. If a mole becomes
greater than ¼ inch is cancerous and E stands for
EVOLVING. High closeness between various sorts of skin
lesions makes a visual appraisal troublesome and prompts
wrong examination. Therefore, an automated system is Fig 3.1 Sample images from Dataset
proposed to ease out the skin lesions classification.

II. OBJECTIVES B. PRE-PROCESSING:

1. To design and develop novel image processing
approached based system for melanoma detection.
2. Segregating the available datasets into benign and
malignant region based on training the network.
3. Segment the region of interest effectively by
removing the noise.
4. Feature extraction for effectively extracting
properties of benign and malignant region.
5. Training Neural Network with variation in layers to
find best efficiency.
The Inception v3 architecture of Convolutional Neural
Network is used to create a model. Image preprocessing is
said to be crucial part of system and can directly influence
accuracy that the model attains. Inception v3 offers many
options for pre-processing stage and also it depends on
whether the model is under training or used for testing.
During training, the image cropping is randomized i.e.,
bounding box is chosen in random manner to select the
region of interest, and then it is resized. This resized image
is flipped and its colors are distorted. Scaling is also
performed.
During testing, the pre-processing includes cropping the are using firebase Google cloud in our project to store and
central region of image and resize it to default i.e., 299x299 retrieve details of the user and hospitals.
size.

C. TRAINING MODEL WITH DATASET:

Dataset consists of two folders: Benign and Malignant.
Benign folder consists of 1799 images of benign melanoma
and malignant folders consists of 1500 images of malignant
melanoma. The 90% of dataset is used for training the
model. The output of the training part is taken in the log file
which contains the output graph and output label. The rest
10% dataset is used for testing the model.

Fig 4.2 Snapshot of scan result

Fig 3.2 CNN Diagram
D. CLASSIFICATION:
Once the model is trained, an image will be sent for
classification to determine whether it is benign or malignant.
Mapping is done between input image and output graph,
whichever class is matching maximum that resultant class
will be the output.
IV. EXPERIMENTAL RESULTS
Firstly, user has to register by entering details like Name,
email, age and city, select the gender. We have mainly used
Tkinter, GUI programming toolkit of python to create the
GUI. The Image in jpeg, jpg or jfif format is selected and
uploaded, this is supported using Python imaging library
(PIL) of python.
Along with the scan result the details of the best skin
specialist of the city that he had entered during the
registration will be displayed in a new window. An email
feature is also implemented. If the user wants these details, it
will be mailed to the user.
Fig 4.3 Snapshot of best skin specialist of that city
V. CONCLUSION
In this paper, various steps in detecting melanoma skin
cancer were discussed. In proposed system the image
preprocessing, segmentation and classification steps are
performed for categorizing skin lesion images into
malignant or benign. This method detects Melanoma faster
than the biopsy method.
As a part of future work the proposed method can include
identification of other skin related diseases.

ACKNOWLEDGMENT

We offer our profound thanks to Mrs. Beena K for her

important and useful ideas during the arranging and
improvement of this undertaking. Her eagerness to give her
time so liberally has been particularly valued.

We are thankful to every one of the teachers of KSIT for

their persistent help and consolation.

Fig 4.1 Snapshot of Registration page

The output will be shown on the GUI whether the patient is

suffering from benign or malignant melanoma. We have REFERENCES
implemented a voice feature that reads out the output. We
[1] Skin Cancer Classification using Deep Learning –
Khalid M. Hosny, Mohamed A. Kassem , and Mohamed
M. Foaud.
[2] SKIN LESION ANALYSIS TOWARD MELANOMA
DETECTION Noel C. F. Codella1†, David Gutman
IEEE 2017
[3] W. Barhoumi and E. Zagrouba, "A Preliminary approach
for the automated recognition of malignant melanoma,"
Image Analysis and Stereology Journal,vol. 23, pp. 121-
135, 2004
[4] G.Maglogiannisand E. P.Zafiropoulos,"Characterization
of digital medical images utilizing support vector
machines," BMC Medical Informatics and Decision
Making, 2004
[5] Nazia Hameed , Kamal Abu Hassan, and M A Hossain
-"A comprehensive survey on image-based computer
aided diagnosis systems for Skin cancer" SKIMA 10th
International Conference,2016.
[6] Joanna Jaworek-Korjakowska & Pawel Kleczek -
"eSkin: Study on the Smartphone Application for Early
Detection of Malignant Melanoma " Wireless
Communications and Mobile Computing Volume, 2018.