2018/05/08 by Andrew Beers, James M. Brown, Beers, Andrew +11 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · Engineering · #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Image Processing Techniques and Applications
paper · pdf · doi:10.48550/arxiv.1805.03144
openalex publication_date 2018/05/08 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Generative adversarial networks (GANs) are a class of unsupervised machine\nlearning algorithms that can produce realistic images from randomly-sampled\nvectors in a multi-dimensional space. Until recently, it was not possible to\ngenerate realistic high-resolution images using GANs, which has limited their\napplicability to medical images that contain biomarkers only detectable at\nnative resolution. Progressive growing of GANs is an approach wherein an image\ngenerator is trained to initially synthesize low resolution synthetic images\n(8x8 pixels), which are then fed to a discriminator that distinguishes these\nsynthetic images from real downsampled images. Additional convolutional layers\nare then iteratively introduced to produce images at twice the previous\nresolution until the desired resolution is reached. In this work, we\ndemonstrate that this approach can produce realistic medical images in two\ndifferent domains; fundus photographs exhibiting vascular pathology associated\nwith retinopathy of prematurity (ROP), and multi-modal magnetic resonance\nimages of glioma. We also show that fine-grained details associated with\npathology, such as retinal vessels or tumor heterogeneity, can be preserved and\nenhanced by including segmentation maps as additional channels. We envisage\nseveral applications of the approach, including image augmentation and\nunsupervised classification of pathology.\n