Pigmented Dermatological Lesions Classification Using Convolutional Neural Networks Ensemble Mediated by Multilayer Perceptron Network

Skin cancer has the highest occurrence rate compared to other types of cancer. This paper presents the development of a Deep Learning model, trained from the Skin Cancer MNIST database (HAM10000). It is able to perform Classification of Pigmented Dermatological Lesions using Convolutional Neural Networks techniques by proposing an ensemble with Multilayer Perceptron Neural Networks. In order to evaluate the Convolutional Networks, the metrics Accuracy, Precision, Revocation and F1-Score were taken into consideration. The ensemble implementation was based on a Grid Search with Cross Validation and evaluated according to Accuracy. The obtained results show the relevance of the research and the consolidation of the techniques used in the development of Artificial Intelligence solutions applied to clinical images analysis. Accuracy and Recall reached 0.93 and Precision and F1-Score 0.92, which is superior performance to specialists and related researches.