Deep Neural Networks for Skin Cancer Classification: Analysis of Melanoma Cancer Data

Abstract

The skin is the largest organ in the human body, serving as its outermost covering. The skin protects the human body from elements and viruses, regulates temperature, and provides cold, heat, and touch sensations. A skin lesion is a type of abnormality in or on the skin. Melanoma skin cancer is the most deadly and deadliest of the skin cancer family. Several researchers have developed noninvasive approaches for detecting skin cancer as technology has advanced. The early detection of a skin lesion is crucial for its treatment. In this study, we introduce a deep neural network for diagnosing skin melanoma in its early stages using Convolutional Neural Network (CNN), Capsule Neural Network (CapsNet), and Gabor Capsule Neural Network (GCN). To train the models, the International Skin Imaging Collaboration (ISIC) melanoma data is used. Prior to deploying deep neural networks, methods such as preprocessing dataset images to remove noise and lighting concerns for better visual information are used. Deep Learning (DL) models are employed to classify the images' melanoma lesions. The performance of the proposed approaches is evaluated using cutting-edge performance metrics, and the results show that the presented method beats state-of-the-art techniques. The models achieve an average accuracy of 90.30% for CNN, 87.90% for CapsNet, and 86.80% for GCN, demonstrating their capability to recognize and segment skin lesions. These developments enable health practitioners to provide more accurate diagnoses and help government healthcare systems with early identification and treatment initiatives.