Deep Learning for Disease Classification in Teledermatology System Using Dermoscopic Skin Images

Abstract

In recent years, the popularity of Deep Learning has surged. Among the most well-known architectures in Deep Learning are Neural Networks, including Convolutional, Recurrent, and Generative Adversarial Networks. Convolutional Neural Networks (CNNs) have become an important architecture for image classification tasks due to their superior accuracy and performance. Skin cancer, the most common form of human cancer, has an extremely high cure rate when detected and treated at an earlier stage. However, automated categorization of skin lesions is challenging due to the fine-grained heterogeneity in their appearance. This study proposed a CancerVisionNet (CNNs) model to predict and categorize seven distinct types of skin lesions. The "Human Against Machine with 10000 training images" (HAM10000) dataset, which contains dermoscopic images, is used in this research to evaluate the proposed method for diagnosing and organizing skin disorders. ReLU as an activation function is employed to handle non-linearity in hidden layers. Our proposed method achieved a higher accuracy (79.6%) than other state-of-the-art methods such as with accuracies of 75.03% and 74.3%. This paper shows the effectiveness of the proposed method in disease classification using dermoscopic skin images in the context of teledermatology.