ABSTRACT
Generative Adversarial Networks (GAN) advocate sophisticated domain-specific data augmentation by making use of a number of GAN techniques, wherein the accuracy of the results derived from synthesizing effectually reduces the problem of data scarcity in medical imaging. GANs generally use a generator and a discriminator model that are trained together in which there is a reproduction of new image and subsequent identification of the original from the generated ones. To optimize on the clinical detection and decisions, GANs are used in conjunction with several deep learning models, where the adversarial learnings are used to train them to improve the accuracy of segmentation. These frameworks can then be generalized for CAD(x) or CAD(e). GANs’s capability to model high-dimensional data, handling missing data, and making provisions for multiple plausible answers in the segment of image synthesis and translation has culminated in frequent usage of this technique in computer aided diagnosis.
