Classification of Non-tumorous Facial Pigmentation Disorders Using Generative Adversarial Networks and Improved SMOTE

The diagnosis of non-tumorous facial pigmentation disorders is crucial since facial pigmentations can serve as a health indicator for other more serious diseases. The computer-based classification of non-tumorous facial pigmentation disorders using images / photographs allows automated diagnosis of such disorders. However, the classification performance of existing methods is still not satisfactory due to the limited real-world images available for research. In this paper, we proposed a novel approach to applying generative adversarial network (GAN) with improved synthetic minority over-sampling technique (Improved SMOTE) to enhance the image dataset with more varieties. With the application of Improved SMOTE, more data is provided to train GAN models. By utilizing the GAN to perform data augmentation, more diverse and effective training images can be generated for developing classification model using deep neural networks via transfer learning. A significant increase in the classification accuracy (>4%) was achieved by the proposed method compared to the state-of-the-art method.