Detección de tumores cutáneos malignos y benignos usando una red neuronal convolucional.

Abstract

The present work focuses on the use of the Inception v3 convolutional neural network architecture and the transfer learning to create a model that can classify dermoscopic images of melanoma (malignant cutaneous tumor) and nevus (benign cutaneous tumor) with a high accuracy. Lacking a large amount of data for training, the data augmentation method is used to provide more information to the classifier in the execution of training and validation tasks. For the training of the last fully connected layer, the mini-batch gradient descent algorithm is used, obtaining a high accuracy value without using too many computational resources, achieving an accuracy of 89,63%. Finally, a web application is used that serves a resource based on the REST architecture to use the generated model by a mobile application.