Classification of skin images with respect to melanoma and nonmelanoma using the deep neural network

Abstract

Melanoma is the most common type of skin cancer. At first, for the diagnosis of melanoma, clinical screening is performed and then diagnosis is made by clinical imaging. It is followed up by dermoscopic analysis, biopsy and histopathological examination. Early diagnosis is important in the treatment of melanoma. Automatic recognition of melanoma from dermoscopy images is a difficult task. Therefore, computer aided systems are recommended to reduce time ,cost and accuracy diagnosis. In this paper, a deep learning-based system is used to classify melanoma in color images taken from dermoscopy devices. With this system, differentiation from previous studies can be done with good accuracy without segmentation step and feature extraction. This system provides a significant advantage in hardware implementation. Because there are no preprocessing and segmentation steps. The International Skin Imaging Collaboration database for the designed system is used and includes 1483 training, 517 test data(ISIC). As a result of the classification of these data, the success rate is reached 86-85%.