Generating digital images of skin diseases based on deep learning

Data generation systems and architectures seek to create new and valuable data with specific sizes and characteristics which are similar to the original data that fit the mechanism of the application used. GANs (generative adversarial network architectures) are a type of generative modeling that uses deep learning methods like convolutional neural networks to generate data, particularly digital images which are used in scientific applications. The goal of this paper is to design and build three architectures of Generative adversarial network (based on convolutional neural networks) to generate three types of digital images of skin diseases by defining two supervised models, the generator model that is trained to generate new digital images and the discrimination model that classifies digital images as real or fake from the field. An intelligent architecture for training the generative model was created, where the two models are trained together in a zero-sum field. Each digital image that was generated differs in its accuracy, size, and dimensions. After applying all the architectures, obtaining and comparing digital images, it turns out that the fewer dimensions of the resulting digital images, the faster the generation processes and the lower the memory costs, but their accuracy is low. The importance of building architecture lies in increasing the images of skin diseases to be used in the data set to perform classification operations for these diseases, as digital images were generated in different sizes, which are 64×64, 128×128, and 512×512, and the new images were obtained with accuracy commensurate with the requirements of this paper. The generated digital images with dimensions 128×128 gave the best results in the accuracy of the resulted images. At the same time, they do not consume much memory, which leads to their processing speed.