A MULTI-STAGE GAN BASED FRAMEWORK FOR THE GENERATION OF HIGH-RESOLUTION MAMMOGRAPHY IMAGES
Rangan Das, Pradip Basak, Utsav Bandyopadhyay Maulik, Saumik Bhattacharya, Ujjwal Maulik · 2020
Generative adversarial networks (GANs) have garnered a lot of attention because of their ability to generate data.This synthetic data generation is crucial for training machine learning models for domains where data is scarcely available, such as medical imaging, where the generation of data is not only expensive but also has privacy concerns.In these cases, such synthetic data can be used for data augmentation, for detecting out-ofdistribution samples or for domain transfer tasks.Generating high-resolution medical images, such as mammograms are challenging because of their detailed features as well as their tissue-specific properties.In this work, we have implemented a multi-stage GAN based framework that can be used to synthesize highresolution mammography images.In the first stage, a low-resolution image is generated from latent noise, while in the second stage, an Enhanced Super Resolution GAN (ESRGAN) is used to upscale the image.