Detection of stages of melanoma using deep learning

Abstract

Human Skin is the most utilized and largest organs next to blood acts as an outer protective covering of the entire body to protect the underlying internal organs from harmful UV rays, dust and pollution. One of the major concerns of such harmful rays is its ability that affects the human skin by mutating the DNA sequence of melanocytes cells which in turn stimulates an uncontrollable rapid growth of infelicitous cells as undesirable peripheral protuberances and in later progressive, prolonged and well-infected stages, it starts to grew inwards as well and reports majority of the deaths due to a skin cancer across the world. The premature detection of the melanoma enables the complete cure of the cancer and it significantly drags down the death rate. Though many research works concentrated on detecting the melanoma, this work narrow down the scope in finding the levels of skin carcinoma by using deep learning methodologies that boosts the accuracy in the detection of melanoma and can give an appropriate treatment with respect to the level of the cancer. In this work, state-of-art techniques like Random Forest, Support Vector Machine, Artificial Neural network along with the proposed fusion-based Deep learning methodology has been experimented. Various performance metric such as Mean Square Error, Peak Signal to Noise ratio for assessing the quality of the pre-processing strategy and Accuracy, Precision and Recall for evaluating the proposed methodology. With the experimental results, it is evident that the Deep learning using the feature-fusion methodology has the accuracy of 97% than other state-of-art techniques. On comparing with recent works done using same methodology, the proposed work has 11% more accuracy than the existing well-known works.