Evaluating the efficacy of GAN-generated synthetic medical images in enhancing deep learning-based diagnosis of rare diseases

Abstract

The leap in medical research is challenged by the scarcity of publicly available data. The insufficient and walled data results in hampering innovative research in this field. The primary objective of this thesis is to compare and evaluate some of the existing Generative Adversarial models and their efficacy on solving this data scarcity issue by generating accurate and close to the real medical images in large volume effectively. This work proposes some effective tweaks and techniques that help immensely in bringing stability to these generative models. Classification models like ResNet50 and Grad-CAM are used to validate the image generation capabilities of these models. The findings of this thesis offer insights into the potential of GAN-generated synthetic medical images to significantly improve the diagnostic capabilities of deep- learning models for rare diseases. As a result, This work not only advances medical imaging and diagnosis but also provides a valuable solution to the persistent issue of data scarcity in the realm of rare medical conditions, ultimately leading to improved patient outcomes and healthcare practices.