Towards to use image mining to classified skin problems - A melanoma case study

Abstract

Data mining (DM) is the area where are discovery patterns and the relationship between data. Thus, depending on the type of data, the process will have different names. In this article is used Image Mining (IM), an area that uses DM techniques to find relationship through classification or clustering, using image data. Specifically, in this paper are expressed algorithms that are trying to predict melanomas using skin lesion and melanoma image’s, avoiding the need for a histologic exam. Furthermore, this avoidance is beneficial in terms of the costs and the level of intrusion that the patient usually suffers. In relation to the solutions, are presented two trained Convolutional Neural Networks (CNN), using packages from Keras and TensorFlow. On the other hand, focusing on the results, the best model was one that only used Keras with an accuracy value of 91% and a loss value of 36% using the testing data. As a conclusion, as a last use case for the model, in the long term it could be used in a preventive way, used to detect if a skin lesion is a melanoma. These benefits are major improvements in current Healthcare techniques. Furthermore, the accuracy of the models, considering only that they only serve as a proof of concept, is considered a “success”.