Image Augmentation for Deep Learning Based Lesion Classification from Skin Images

Abstract

Skin lesion classification based on deep learning models, which are data-hungry, is a challenging issue because of the shortage of annotated images and unbalanced classes in image sets. The lack of sufficient number of labeled data or class unbalancing in image sets lead to overfitting problems affecting robustness and generalization ability of network models. Image augmentation is an efficient approach to deal with this issue using existing images more efficiently. In addition to image augmentations, various solutions have been developed in the literature to solve the overfitting problem and to obtain well-generalizing network models. However, there is not a clear way how the most appropriate solution should be selected. Therefore, in this paper, those alternative solutions and image augmentations applied recently in deep learning based skin lesion classifications are presented.