Data augmentation importance for classification of skin lesions via deep learning

Abstract

Melanoma is a fatal type of cancer which is mostly curable if detected in early stages. Various machine learning algorithms are used for distinguishing benign lesions from malignant such as deep learning. To obtain successful result from deep learning, large and quality training data set is essential. But, existing data sets maybe insufficient for training a deep learning network. Building a powerful classifier from insufficient data, data augmentation methods are useful. In this article the same network trained with augmented skin lesion images and non- augmented skin lesion images for detecting malignant skin lesions. When compered results, it has been seen that the network using augmented data for training has achieved better results than training with non-augmented data.