Skin cancer disease images classification using deep learning solutions

Abstract

Skin cancer is a type of dangerous disease, and early detection is necessary to increases the survival rate. In recent years, deep learning models applied to computerized skin cancer discovery has become a standard. These models can improve their performance by being able to access more data and its main task is to the classification of images. This task is exceptionally valuable in the field of medicine, it has the ability to assist doctors and specialists to make the right decision and diagnose the patient’s condition with high accuracy. In this paper, a deep learning network has been selected and trained by the author for the analysis of more than 24,000 skin cancer images by convolutional neural network (ConvNet) model applying with three architectures (InceptionV3, ResNet, and VGG19) with many parameters to identify the best architectures in the classification of these images and getting extremely acceptable results; and classifying the cancer type as benign or malignant with high accuracy. The dataset contains high-resolution images obtained from the ISIC archive between 2019 and 2020. After all the tests were done, the best architecture is InceptionV3. This architecture has achieved a diagnostic accuracy of approximately 86.90%, precision of 87.47%, sensitivity of 86.14%, and the specificity of 87.66%.