Skin cancer diagnosis using deep learning algorithms

Skin cancer is a type of cancer that can be fatal if not diagnosed early. Automatic diagnosis of skin cancer with artificial intelligence plays a critical role in reducing the prevalence of skin cancer and saving lives by enabling early diagnosis and treatment. This study summarizes research conducted to classify skin cancer tumor types using the ISIC2019 dataset. According to the research results, accuracy of 98.28% in ResNet18, 93.10% in ResNet50, and 98.28% in VGG16 were achieved in the Melanoma-AC classification. Similar success rates have been reported for other tumor types. The study consists of five main sections. In the first part, an introduction and literature summary are presented, and in the second part, general information about skin cancer and definitions of the tumor types in the data set are given. In the third section, information about the ResNet18, ResNet50 and VGG16 architectures, which are the artificial intelligence algorithms used, is presented, and their layer structures are explained in detail. In the fourth section, the methodology and analysis results are explained. In the fifth section, the obtained results are discussed, and the performances of the network architectures are evaluated. The results show that high accuracy rates are achieved in classifying skin cancer tumor types using CNN. Particularly noteworthy is the 100% accuracy in the classification between melanoma and vascular lesions. Although the lowest success rates were obtained in the Melanoma-BCC and Melanoma-Nevus classifications, even these results show that a useful classification system has been developed for doctors. When network architectures are compared, ResNet50 is seen to be more successful, but it requires more processing and has a longer training time. ResNet18 and VGG16 have lower classification success. VGG16 showed better performance despite having fewer layers and a simpler architecture. This study highlights that ESA is an effective option for skin cancer classification and can be used as a system that can assist doctors in the future.