Detecção de lesões de câncer de pele utilizando análise de componentes independentes e análise discriminante linear

Abstract

Skin cancer has become the most frequent in the Brazilian population. Excessive exposure to solar radiation is the major risk factor. White people have this risk increase. The rates of incidence and morbidity of skin cancer have been increasing, so the early diagnosis of this disease is the simplest method to reduce these rates. This work aims to present a CAD method based on image processing for detection of melanoma and non-melanoma skin cancer using independent component analysis for feature extraction, the algorithm of maximum relevance and minimum redundancy for dimensionality reduction and analysis Linear discriminant for the classification of images. The ability of the method was evaluated by the cross-validation technique, and reached 100% accuracy, 100% sensitivity and 100% specificity when analyzing all samples.