Evaluation of Melanoma Diagnosis using Deep Features

Abstract

Melanoma is the most lethal type of skin cancer when compared to others, but patients have high recovery rates if the disease is discovered in its early stages. Several approaches to automatic detection and diagnosis have been explored by different authors, using pattern recognition and machine learning techniques. This work has as main objective to evaluate the main architectures of Convolutional Neural Networks for the task of diagnosis of melanoma skin cancer. Different classifiers were tested with the characteristic vectors extracted by the networks on the dataset PH2 of dermatoscopic images. The results were promising when combining the VGG19 network and Logistic Regression, with accuracy of 92.5% and 85.71% of precision.