Classification of automatic skin lesions from dermoscopic images utilizing deep learning

Abstract

Skin cancers are the most incidental in Brazil. Thousands of Brazilians are diagnosed annually with the disease, which occurs due to the abnormal growth of the cells that make up the skin and, therefore, can give rise to several types of skin cancer, being divided into two types melanoma and nonmelanoma. Skin cancer, which represents respectively 25% of the malignant tumors and about 30% of the cases of cancer in Brazil, being the majority due to the excessive exposure to the sun’s ultraviolet rays. Skin tumors are usually perceived more efficiently, and when diagnosed early, they are more likely to heal. The present study is relies on the development of an algorithm based on Deep Learning for the recognition of tumors in skin images. The AlexNet, which is a Deep Learning architecture is modified to attending our classification problem. The experiments are conducted through 1400 and 2400 images, after twice training with different optimizer, SGD is the better optimizer with 99.79% of accuracy and 0.0120% of loss in training, for the scenery of 2400 images