This document summarizes research on using generative adversarial networks (GANs) to augment medical image datasets. Specifically, it discusses using CycleGAN to generate additional images of rare disease classes, like tubular adenoma and sessile serrated adenoma, by transforming normal images. Evaluations found generated images were realistically classified by pathologists and inclusion of generated data improved classification performance of models trained on real data. In summary, GANs show promise for addressing limited rare class image data by generating synthetic augmented images.