Early detection of skin cancer using deep learning architectures: resnet-101 and inception-v3

Abstract

Skin cancer is one of the most prevalently seen cancer type in human beings. Skin cancer occurs due to the uncontrollable growing of mutations taking place in DNAs owing to some reasons. Recognizing the cancer in early stages could increase the chance of a successful treatment. Nowadays, computer aided diagnosis applications are used almost at every field. One of the mostly used areas is health sector. Biomedical datasets are created by saving the data of illness people in computers. Our goal is to obtain an effective way for early diagnosis of skin cancer by classifying our dataset images as benign or malignant. Our dataset consists of 2437 training images, 660 test images and lastly 200 validation images. ResNet-101 and Inception-v3 deep learning architectures are used for the classification task. Once the acquired results are examined, an accuracy rate of 84.09% is get in ResNet-101 architecture, and an accuracy rate of 87.42% is get in Inception-v3 architecture.