AUTOMATIC SKIN CANCER DETECTION IN DERMOSCOPY IMAGES BASED ON ENSEMBLE LIGHTWEIGHT DEEP LEARNING NETWORK

Abstract: Skin cancer affects 30,000 people each year, according to the World Cancer Research Fund. Most frequently, skin exposed to the sun gets skin cancer, an abnormal development of skin cells. But this typical sort of cancer can also develop on parts of your skin that aren't usually exposed to sunlight. Melanoma and Benign are the two main kinds of skin cancer. It is currently very difficult to automatically diagnose different skin lesion disorders using medical dermoscopy images. In this study, a cascading innovative deep learning network-based integrated model for segmenting skin lesion boundaries and classifying skin lesions is proposed. In the first stage, the boundaries of skin lesions are segmented from dermoscopy pictures using a unique full resolution convolutional network (FrCN). Following that, a deep residual network is fed with the segmented lesions to classify them.

Keywords: Skin cancer; Deep learning ; Dermoscopy ; Full resolution convolution network