Medical Image Generation Techniques for Data Augmentation: Disc-VAE versus GAN

AI algorithms can analyze medical images, such as X-rays, MRIs, and CT scans, assisting doctors in detecting diseases and making more accurate diagnoses. However, large, diverse datasets of medical images are critical for training robust and effective AI models. Furthermore, acquiring and annotating these datasets poses considerable challenges due to factors such as cost, privacy concerns, and data scarcity. To address these challenges, data augmentation techniques, particularly synthetic image generation, have become more popular in recent years. Traditional augmentation methods such as rotation, scaling, and cropping are commonly used but may not capture all the characteristics of medical images.This paper explores various techniques for generating synthetic medical images to enhance AI models. We examine recent methodologies of data augmentation employed for generating synthetic medical images. Additionally, we highlight the importance of synthetic images in improving the performance and generalization capabilities of machine learning algorithms in the medical field. This is achieved by adding the discriminator into the VAE, Disc-VAE, and comparing it with GAN through the quantitative metrics (SSIM, PSNR, and Accuracy).