Skin Cancer Classification Using Convolutional Neural Networks

Abstract

Skin cancer is the out-of-control growth of abnormal cells in the epidermis. It is usually caused by damage or mutations in the DNA. This causes the skin cells to grow and multiply rapidly and out of control, which forms tumors in the skin. Deep learning is a type of machine learning that—through many layers of artificial neurons— learns high-level features from a high-dimensional input. In image processing, the lower layers of a neural network may identify simple features such as edges, while later layers may learn complex concepts such as digits or letters. The testing and training data was downloaded from the ISIC 2019 Challenge which is a database with images of skin cancer images ranging in various types. Training images were run through the three convolutional layers with ReLU activations, each followed by a max-pooling layer. Finally, they were passed through three dense layers and a softmax layer activation. Here, we trained a 2D CNN on images of skin lesions to classify benign patterns and different types of skin cancers. The model achieved reasonably high accuracy on the most common kinds of skin lesions. These findings suggest that with the right data and modeling, a neural network can be used to limit the number of doctor visits people need to make to determine if they have skin cancer or not.