Computer Vision-Based Automated Diagnosis for Skin Cancer Detection

With an increased life expectancy and ageing population, noninvasive diagnostic and clinical tools must be developed that meet safety standards. The efficiency of a healthcare system could be significantly improved if methods, tools and devices are developed that can automate and collect data continuously. Several devices and techniques for monitoring vital signs, such as temperature, blood oxygen saturation, heart rate and breathing rate, have been developed in recent years. However, there are many challenges associated with developing such tools and techniques, such as sensitivity, specificity, reliability, durability and efficiency. From the wider perspective of noninvasive healthcare monitoring, skin cancer is by far the most common type of cancer, and it has been reported to have significantly increased over the past decade, with more than 1.5 million cases of both nonmelanoma and melanoma. With this increase in the incidence of skin cancer, several techniques have been investigated. Although dermoscopy is a powerful technique for the diagnosis of late-stage melanoma, early-stage identification of melanoma is crucial for preventing metastasis, which can significantly improve the survival rate and reduce the cost of treatment. Artificial neural network (ANN)-based techniques such as deep learning have gained popularity in recent years and play a significant role in skin cancer detection. Therefore, it is important to analyse the research findings in the wider context of machine learning. This chapter will critically review and present a case study to explore the efficiency and accuracy of different neural network-based techniques. A focused retrospective narrative will be provided for the readers in selecting appropriate diagnostic tools that could be adopted for both fine-tuning the parameters and image classification for skin cancer detection.