Classification of Skin Phenotype: Melanoma Skin Cancer

Abstract

Skin cancer (skin phenotype) is most common cancer in United State of America (USA). Skin cancer can affect anyone, regardless of skin color, race, gender, and age. The characteristics of skin phenotype of melanoma lesion has an arbitrary shape, size, uneven and rough edge, and cannot be divided in half. Further, it is a leading cause of deaths worldwide. Every year, more than 5 million patients are newly diagnosed in USA. The deadliest and serious form of skin cancer is called melanoma. The diagnosis of melanoma has been done by visual examination and manual techniques by skilled doctors. It is time consuming process and highly prone to error. The skin images captured by dermoscopy eliminates the surface reflection of skin and gives better visualization of deeper levels of skin. In spite of these, image of skin lesion has many artifacts, noises, complex nature of lesion structure. Due to these complex natures of images, the border detection, feature extraction, and classification process is a complex problem. In order to identify and predict melanoma in early stage, there is need to classify images using better classification and prediction algorithms. Therefore, there is need to make an efficient, effective, and accurate melanoma identification, classification, and prediction such that it may be identified and classified in very early stage. The goal of this poster is to review and analyze the various classification deep learning algorithms - Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) - on images of skin lesions on each one of those and test with publicly available International Skin Imaging Collaboration (ISIC) archive large data sets. Also, ISIC raw datasets will be preprocessed and resized to make the data compatible to algorithms. Moreover, the performance of these algorithms will be measures and compared using five parameters including ROC.