Comparative Analysis of Melanoma Classification Using Deep Learning Techniques on Dermoscopy Images

Abstract

Melanoma is developed due to disturbances in the melanocytes of the skin cells. Every human skin color is defined by melanin, which is produced by melanocytes. Around the globe, the melanoma probability percentage age intervals are from 70 and older. Melanoma is a lethal form of cancer and can easily spread to other parts of the body. It needs to be detected and treated early to avoid mortality. Early diagnosis can be made by an automated diagnosis system to help clinicians for larger populations. In the proposed system, the input images are taken from Med Node, PH2, and HAM10000 Kaggle and given to the pre-trained architectures such as AlexNet, Vgg-16, ResNet50, Inception V3, and GoogleNet. The performance is analyzed using accuracy (AC), sensitivity (SE), specificity (SP), positive predictive value (PPV), and negative predictive value (NPV). Among all the architectures, InceptionV3 obtained the best accuracy of 97.1%, 97.2%, and 96.2% for the MedNode, PH2, and HAM10000 Kaggle datasets, respectively, in melanoma identification.