Skin cancer classification in dermoscopy images using convolutional neural network
S. Alagu, K. Bhoopathy Bagan
Abstract
Skin cancer is a disease caused by an excessive growth of skin cells in abnormal manner. It affects all humans with different skin tones. Sun exposed areas like face, hands and legs are widely affected. Skin cancer is highly treatable and curable when it is diagnosed at the early stage. Basal, squamous and melanoma are various types of skin cancer. Melanoma is the most harmful cancer than others. Skin lesions like melanoma and non-melanoma cells look similar and it is very difficult for human being to discriminate. The most reliable system is required for identifying the skin cancer cells with greater accuracy. The main objective of proposed research work is to identify the melanoma cells using convolutional neural network architecture incorporated with optimization techniques. The cancer cells images are obtained from online database. Data augmentation is performed for increasing the number of images. After resizing all the images into uniform size, the images are given as input to convolutional neural network. The DenseNet architecture with transfer learning is adopted for skin cancer classification. RMSProp, Adam and Stochastic Gradient Descent methods are utilized for better results. It is found that the classification accuracy of 95% is obtained using CNN with Stochastic Gradient Descent optimization technique.