Abstract

Skin cancer has become increasingly prevalent in several countries across the world. The steady rise of the disease in the UK has become a healthcare burden on both medical practitioners and patients. To reduce mortality rates, it is necessary to identify and diagnose the disease early and accurately before it becomes fatal.

The main types of skin cancer are squamous cell carcinoma, basal cell carcinoma and melanoma. This dissertation is focused on the most severe type of skin cancer which is melanoma. This cancer begins in the melanin, making cells melanocytes, and is likely to spread to other body parts and tissues if not identified immediately. There are current diagnostic practices used to identify if a skin lesion is malignant or benign - these procedures are mostly based on visual perception from a medical specialist. However, this project highlights the flaws in these procedures - they are unreliable and several risks could be incurred from them. This project would also consider the use of smartphone applications (apps) in appropriately identifying skin lesions and would analyse the reliability of currently existing skin cancer detection smartphone apps.

Outstanding performance results have been achieved from applying deep learning to the medical field and this has aided in improving diagnostic accuracy. In this thesis, some deep learning algorithms which would aid in alleviating the problems associated with current melanoma detection practices would be discussed. These algorithms include Convolution Neural Networks (CNN), Support Vector Machines (SVM), Deep Ensemble for Recognition of Melanoma (DERM), and Artificial Neural Networks (ANN). A suitable model would be identified, to facilitate effective and early diagnosis of the disease, thereby decreasing health and financial burdens on patients and improving survival rate.

Artificial Intelligence System to Improve Melanoma Detection