Dermoscopic Images Classification Using Pretrained VGG-16 and ResNet-50 Models

Abstract

Skin cancer is the uncontrolled proliferation of skin’s cells due to overexposure to ultraviolet (UV) radiation. It can be classified into benign and malignant lesions, the last ones are deadly and have the potential to metastasize. For example, melanoma, a type of malignant lesion, can spread extensively and is fatal. For this reason, early detection is crucial as it helps avoid the need for time-consuming biopsies and reduces the impact of heavy treatments for patients. This research introduces an automated skin cancer classification system using ISIC archive dataset (International Skin Imaging Collaboration) from Kaggle. In this study, we focus on deep learning, particularly transfer learning, when we use VGG-16 (Visual Geometry Group) and ResNet-50 model (Residual Network) for the classification of dermoscopic images. The VGG-16 model achieves an accuracy of 91.20%, while ResNet50 model achieves an accuracy of 92.80%. In conclusion, our study presents encouraging results that could assist dermatologists in their diagnosis.