High-resolution medical image synthesis using progressively grown generative adversarial networks

Generative adversarial networks (GANs) are a class of unsupervised machine algorithms that can produce realistic images from randomly-sampled in a multi-dimensional space. Until recently, it was not possible to realistic high-resolution images using GANs, which has limited their to medical images that contain biomarkers only detectable at resolution. Progressive growing of GANs is an approach wherein an image is trained to initially synthesize low resolution synthetic images(8x8 pixels), which are then fed to a discriminator that distinguishes these images from real downsampled images. Additional convolutional layers then iteratively introduced to produce images at twice the previous until the desired resolution is reached. In this work, we that this approach can produce realistic medical images in two domains; fundus photographs exhibiting vascular pathology associated retinopathy of prematurity (ROP), and multi-modal magnetic resonance of glioma. We also show that fine-grained details associated with, such as retinal vessels or tumor heterogeneity, can be preserved and by including segmentation maps as additional channels. We envisage applications of the approach, including image augmentation and classification of pathology.