Melanoma Detection From Images of Moles Using Neural Networks

Abstract

Moles can be analyzed for certain characteristics to determine whether the mole is malignant or benign and if a patient is at risk for melanoma, a type of skin cancer. A standard multilayer perceptron and convolutional neural network are used to investigate whether machine learning can serve as a substitution for a human examination of such moles. Both neural networks were developed using the Tensor Flow python module, and images of moles were fed as training points and test points to be classified. Both types of networks were able to outperform random guessing, with accuracies of 0.82 and 0.84 for the multilayer perceptron and convolutional neural network respectively.