Melanoma Skin Cancer Detection from Dermoscopic Images using Computer Vision

Abstract

Skin cancer is a form of cancer that is most common among Caucasians and is rapidly increasing year by year. Melanoma is the most dangerous type of skin cancer making it the world’s 17th most common cancer type which is 1.8% of all cancer patients globally. It is vital to detect Melanoma at its early stages to cure the patient without letting the cancer grow further. The proposed solution is to develop a system that can detect Melanoma by analysing the dermoscopic images while extracting Asymmetry, Border Irregularity, Colour, Diameter (ABCD) features and other salient dermoscopic features which are more often visible in Melanoma skin lesions. Different classification methods are also being evaluated to identify which classifier works best with the dermoscopic features extracted in the feature extraction stage so that the highest accuracy could be obtained when diagnosing Melanoma. Therefore, when a dermoscopic image is given to this proposed system it will output whether the patient is diagnosed with Melanoma along with the confidence level of the diagnosis result at which the model performed the diagnosis based on the various dermoscopic features extracted from the lesion.