Detection of skin cancer Melanoma through

Computer Vision.
Wilson F. Cueva, F. Muoz, G. Vsquez., G. Delgado

Universidad del Azuay

Escuela de Ingeniera Electrnica
Summary
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Summary
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Introduction
What is melanoma?
a cancer of the skin
90% of melanoma cases can be linked to exposure to ultraviolet
(UV) rays.
Dermatologist expert, range of 75 and 84%
Summary
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Methodology.
ABCD of melanoma
Asymmetry of the form: one-half of the mole is different from the other.
Borders: blurry, irregular, festooned, diffuse or imprecise.
Color: varied in the same place or multiple colors on the same mole as red, white, black, brown.
Diameter: greater than 6 millimeters.

Acquisition and Processing

Feature extraction and classification of the injury.
Common mole, not common or melanoma.

Classification algorithm / Neural Network.

Summary
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Acquisition and Processing.
DataBase ADDI (Automatic computer-based Diagnosis system for
Dermoscopy Images).
200 moles: common (80), not common (80) and melanomas (40).

a) Common mole b) Not Common c) Melanoma

Summary
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Feature extraction.
Asymmetry.
Mumford-Shah method.
Feature extraction.
Edge.
Harris- Stephens algorithm.
Feature extraction.
Color.
RGB to HSV.
Common Mole: brown, cinnamon or black.
Presence of colors such as white, brown or the mixture of uniform colors
determines if the mole is common, not common or melanoma.
Feature extraction.
Diameter.
More than 6mm.
Summary
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Neural Network for classification.
- Features:
- Feed forward Back propagation.
- 4 inputs (Asymmetry, Edge, Color and Diameter).
- 1 output, can be classified into three forms (Common, Non-Common and
Melanoma).
- Hidden layer of 50 neurons.
- Activation function: pureline (hidden layer) and sigmoid (output layer).

- Training.
- 30 images taken from the database, 10 of each type: common, not common,
and melanoma.
Index
Introduction.
Methodology.
Acquisition and Processing.
Feature extraction and classification of the injury.
Neural Network for classification.
Results.
Results.
Results.
Results.
Results.
Conclusions.
Provide a second opinion for the diagnosis of this disease,
due to the analysis of the ABCD, giving a high degree of
reliability.

After an analysis of 201 images in the algorithm developed

a performance of 97.51% was obtained; if we compare a
doctor (75 to 84 %), the system achieves a high degree of
efficiency.
Future Work.
New Data Bases.
A Cross validation.
Support Vector Machine vs Neural Networks.
Medical Validation in Hospital del Rio.
ING. GABRIEL DELGADO
O.

gabrieldelgado@uazuay.edu.ec

THANKS FOR YOUR ATTENTION