Skin cancer classification using Progressive Growing of Generative Adversarial Network

Skin cancer detection has become a complex task for medical professionals worldwide. Early diagnosis of carcinogens is important for treating skin cancer and reducing mortality rates, but it still remains a daunting task The models available for effective diagnosis of patients provide less accuracy due to data imbalance. Presently, Generative Adversarial Networks are being utilized but the issue of variations within and between different categories of images is still prevalent. This problem arises due to the shortage of data. This study proposes the use of Progressive Growing of Generative Adversarial Network that aims at producing realistic synthetic images using PAD-UFES 20 dataset that consists of non-dermoscopic images. It is further integrated with classifiers based on convolutional networks for better results. This way model over-fitting and data imbalance is prevented, and diagnosis accuracy is enhanced. This study further aims at producing images with better resolution as well as increasing the classification accuracy between cancerous and non-cancerous skin lesions.