CLASSIFICATION OF SKIN CANCER BASED ON DEEP LEARNING USING CONVOLUTIONAL NEURAL NETWORKS – OPPORTUNITIES AND VULNERABILITIES A SYSTEMATIC REVIEW

Abstract

Convolutional Neural Networks (CNNs) have outperformed dermatologists in the classification of skin lesions related to skin cancer, potentially saving lives through earlier diagnosis. By just installing an app on their mobile devices, people will be able to self-diagnose their cancer. By the end of 2021[28], 6.3 billion people are expected to have used the subscriptions to diagnose themselves with skin cancer. This study shows its findings after reviewing a large number of research articles on CNN-based skin lesion classification. Thanks to recent advances in machine learning algorithms, the rate at which skin lesions are erroneously identified has decreased as compared to dermatologist categorisation. This study looks at the approaches that have been taken, the effectiveness of those approaches, and the development of CNN in the successful classification of skin cancer subtypes. While deep learning with CNN gives advantages over a dermatologist, it also has certain disadvantages when misclassifying photos depending on symptoms and criteria. We also address these weaknesses in this overview research. We searched the Science Direct, PubMed, Elsevier, Web of Science, and Google Scholar databases for published original research publications. From the web publications we looked for, we selected articles with sufficient data and information about the authors' study and created an overview of the authors' approaches and methodologies. There is currently a lack of review literature addressing the merits and drawbacks of applying deep learning to the classification of skin cancer. Advances in deep learning and machine learning technology can eliminate human error and prevent errors and classifications. Along with their limitations, we will also discuss the benefits of using CNNs for deep learning.