Pix2pix Conditional Generative Adversarial Networks for Scheimpflug Camera Color-coded Corneal Tomography Image Generation

Computer-aided medical image analysis has advantages, but requires large amounts of training data, which are scarce and costly to obtain, are subject to privacy concerns, and are often highly imbalanced, with over-representation of common conditions and poor representation of rare conditions. Consequently, some methods have been proposed to generate artificial medical images using generative adversarial networks (GANs). Computer aided diagnosis of keratoconus is an emerging research field that may benefit greatly from medical image synthesis, which can affordably provide an arbitrary number of sufficiently diverse synthetic images that mimic real Pentacam images. A new conditional GAN, the pix2pix cGAN, has not been used in this context to date. Here, investigators will assess the efficacy of a cGAN implementing pix2pix image translation for image synthesis of color-coded Pentacam 4-map refractive displays of clinical and subclinical keratoconus as well as normal corneas.