Challenges and Imperatives of Deep Learning Approaches for Detection of Melanoma: A Review

Abstract

Recently, melanoma became one of the deadliest forms of skin cancer due to ultraviolet rays. The diagnosis of melanoma is very crucial if it is not identified in the early stages and later on, in the advanced stages, it affects the other organs of the body, too. Earlier identification of melanoma plays a major role in the survival chances of a human. The manual detection of tumor thickness is a very difficult task so dermoscopy is used to measure the thickness of the tumor which is a non-invasive method. Computer-aided diagnosis is one of the greatest evolutions in the medical sector, this system helps the doctors for the automated diagnosis of the disease because it improves accurate disease detection. In the world of digital images, some phases are required to remove the artifacts for achieving the best accurate diagnosis results such as the acquisition of an image, pre-processing, segmentation, feature selection, extraction and finally classification phase. This paper mainly focuses on the various deep learning techniques like convolutional neural networks, recurrent neural networks, You Only Look Once for the purpose of classification and prediction of the melanoma and is also focuses on the other variant of melanomas, i.e. ocular melanoma and mucosal melanoma because it is not a matter where the melanoma starts in the body.