ESTUDO COMPARATIVO SOBRE CLASSIFICAÇÃO DE LESÕES DE PELE COM RNA-MLP E SVM

Resumo

Technological evolution, evidenced in recent years, has contributed significantly to both
Artificial Intelligence (AI) and Digital Image Processing and Analysis (DIP) areas. Several
computational methodologies can be found in the specialized literature, with different applications, such
as the classification of skin lesions from dermatoscopic images. Although the initial analysis of skin
lesions is based on a set of visual rules known as the ABCD rule (Asymmetry, Borders, Color and
Diameter), the performance of this visual analysis is influenced by factors such as variation of
illumination during capture of the image, the presence of artifacts that cause noise, and the visual
fatigue of the specialist during image analysis. A mistaken initial analysis may lead to delays in the
elaboration of an adequate treatment plan, affecting the effectiveness of this treatment. The
interdisciplinarity between AI and DIP has produced good results when the algorithms are intended for
specific applications and use a priori knowledge based on the problem addressed. In this context, this
work presents a comparative study between the classification methodologies based on Artificial Neural
Networks (ANN) and Support Vector Machine (SVM), using the existing knowledge about the visual
evaluation of skin lesions as knowledge base a priori to train these classifiers. In general terms, it is
expected to better understand the performance of these two AI techniques when applied to classify
skin lesions in malignant and non-malignant, evaluating the performance of each methodology. To
perform this work we used the Weka tool, with the implementation of Artificial Neural Network
(Multilayer Perceptron) and Support Vector Machine (Sequential Minimal Optimization). The dataset
used has 200 examples and 14 attributes. The methodology used was Stratified Cross-Validation in 10
parts. The parameters of each algorithm were defined as standard in the tool. The results obtained
were promising, showing the relevance of the AI algorithms to skin lesion classification and the
relevant features of the Weka to improve the quality of the classification modeling solutions.